250 121 4MB
English Pages 384 [385] Year 2007
Aging of the Genome
Cover: atomic force microscopy (AFM) images of the Orc4 subunit of origin recognition complex (blue-yellow sphere) bound to the DNA replication origin (green strand), from fission yeast, Schizosaccharomyces pombe. The images were acquired using tapping-mode AFM in air (Gaczynska et al., Proc. Natl. Acad. Sci. USA 2004, 101, 17952–17957). The illustration is composed from two zoomed-in images: in the left image 1 cm corresponds to approximately 1 nm, and in the right image 1 cm is 5 nm. The height scale is represented by a false color palette, from blue (about 10 nm) through yellow and green to black (background, 0 nm).
Aging of the Genome The dual role of DNA in life and death Jan Vijg Buck Institute for Age Research, Novato, CA, USA
3
3 Great Clarendon Street, Oxford OX2 6DP Oxford University Press is a department of the University of Oxford. It furthers the University’s objective of excellence in research, scholarship, and education by publishing worldwide in Oxford New York Auckland Cape Town Dar es Salaam Hong Kong Karachi Kuala Lumpur Madrid Melbourne Mexico City Nairobi New Delhi Shanghai Taipei Toronto With offices in Argentina Austria Brazil Chile Czech Republic France Greece Guatemala Hungary Italy Japan Poland Portugal Singapore South Korea Switzerland Thailand Turkey Ukraine Vietnam Oxford is a registered trade mark of Oxford University Press in the UK and in certain other countries Published in the United States by Oxford University Press Inc., New York © Jan Vijg, 2007 The moral rights of the author have been asserted Database right Oxford University Press (maker) First published 2007 All rights reserved. No part of this publication may be reproduced, stored in a retrieval system, or transmitted, in any form or by any means, without the prior permission in writing of Oxford University Press, or as expressly permitted by law, or under terms agreed with the appropriate reprographics rights organization. Enquiries concerning reproduction outside the scope of the above should be sent to the Rights Department, Oxford University Press, at the address above You must not circulate this book in any other binding or cover and you must impose the same condition on any acquirer British Library Cataloguing in Publication Data Data available Library of Congress Cataloging in Publication Data Data available Typeset by Newgen Imaging Systems (P) Ltd., Chennai, India Printed in Great Britain on acid-free paper by Antony Rowe Ltd., Chippenham ISBN 978–0–19–856922–0 978–0–19–856923–7 (Pbk.) 10 9 8 7 6 5 4 3 2 1
■ P R E FAC E
Science is a major force in the introduction of new ideas and information in society. This has not always been so. After a hesitant beginning in the high middle ages, science definitely took off during the late Renaissance as a competing force with religion and began to capture the hearts and minds of many people. While originally motivated by the desire to know life, how it originated, and how it could be extended, science was soon absorbed by the prosaicism of the Industrial Revolution in the eighteenth and nineteenth centuries. From then on science was subject to practical purposes such as industrial manufacture, environmental control, and fighting human disease. As such, science was generally accepted by the general population. Meanwhile, the quest for the origin of life, of who we are, how we live, and how we die never expired and eventually resulted in a remarkably clear picture that is now generally adopted by the more enlightened in society. Ironically, this insight is highly controversial in society as a whole and not accepted at all by a large fraction (probably the vast majority) of the world population. Indeed, Darwin is as controversial now as in the nineteenth century. Meanwhile, biology has come to dominate the science of the twenty-first century and it is no wonder that again, as in the seventeenth century, it is the limit to life that takes hold of the minds of many of our best thinkers. To some extent we have come full circle. The question is again whether we can beat the aging process and disassemble the roadblocks to immortality, this time through the accomplishments of the new biology. Can modern science succeed where hermeticism failed? To know whether it is possible to prevent or cure aging we need to know what it is that makes us lose our vigor, causes disease, and finally, inescapably, leads to death. This book is a recapitulation of one of the oldest and arguably the most consistent theories of how we age. First formulated in the 1950s, the somatic mutation theory explains aging as a gradual accumulation of random alterations in the DNA of the genome in the cells of our body. This theory has proved to be remarkably robust and is compatible with the other major theory of aging that does not die: the free-radical theory of aging. Whereas the latter provides a logical explanation for where most of life’s wear and tear comes from, the somatic mutation theory explains how this can result in physiological decline and increased disease. Or does it? Based on what we now know about the genome, ours as well as those of many other species, how the information it contains is maintained as part of its structural characteristics, and how this information is retrieved and translated into function, is it still reasonable
vi
PREFACE
to see this as the main cause of aging in a time when most of us are convinced that the process is multifactorial and must have many causes? Is it possible that the inherent instability of our genomes is not only responsible for the increased chance of getting cancer in old age, but also in some way has an adverse effect on cell function, results in reduced organ capacity, and causes a variety of physiological changes, as well as such diseases as cardiovascular disease, neurodegenerative disorders, and diabetes? Finally, what are the implications of such a stochastic, molecular basis of aging for all those strategies that are now being designed to keep us alive and healthy a bit longer and possibly forever? This and more will be discussed in this book. I have not been shy to include many results obtained in my own laboratory, but a book of this kind depends heavily on other people’s research and other people’s writing. I have tried to acknowledge this great debt to others as much as I could and there are of course the references. Nevertheless, I am afraid that a substantial portion of what I read in some publication, website, or newspaper, not to mention elements picked up during scientific conferences or learned from some of my colleagues, is not properly acknowledged. I apologize for that in advance and would like to hear about it if at all possible. I am heavily indebted to some of my colleagues for their critical comments on earlier drafts of the different chapters. More specifically, I would like to thank Judy Campisi (Berkeley, CA, USA) and Steve Austad (San Antonio, TX, USA) for their comments on Chapter 1, Steve Austad (San Antonio, TX, USA) and Gordon Lithgow (Novato, CA, USA) for their comments on Chapter 2, Tom Boyer (San Antonio, TX, USA) for comments on Chapter 3, Judy Campisi and Jan Hoeijmakers (Rotterdam, The Netherlands) for comments on Chapter 4, Paul Hasty (San Antonio, TX, USA) for comments on Chapter 5 (which is based on a joint publication), Peter Stambrook (Cincinnati, OH, USA) and Martijn Dollé (Bilthoven, The Netherlands) for comments on Chapter 6, George Martin (Seattle, WA, USA), Huber Warner (St. Paul, MN, USA), and my wife, Claudia Gravekamp (San Francisco, CA, USA), for their comments on Chapter 7, and Huber Warner and Aubrey de Grey (Cambridge, UK) for comments on Chapter 8. I am extremely grateful to my friend and colleague,Yousin Suh (San Antonio, TX, USA), for critically reading the entire manuscript and her many useful comments. Thanks to her helpful input at a very early stage I have been able to find the right direction. I thank the members of my laboratory, now and in the past, for sharing their results with me, for all their hard work and their flexibility in dealing with my often unreasonable demands. I am especially grateful to Jan Gossen and Martijn Dollé, perhaps the best scientists who came from my laboratory and superb scholars in their own right, and to Brent Calder for making many of the figures and for always being ready to help me out during the preparation of the manuscript. Finally, I would like to thank the people of Oxford University Press, especially Nik Prowse for his careful editing and many useful suggestions for improvements, and Stefanie Gehrig and Ian Sherman for their frequent advice during the preparation of the
PREFACE
vii
manuscript. I am also grateful to the anonymous reviewers of the original book proposal for their many useful suggestions, and to Maria Gaczynska and Pawel Osmulski (University of Texas Health Science Center) for contributing the cover illustration. And last, but not least, I thank my wife, Claudia Gravekamp, for her patience and nonabating support during the course of this work.
This page intentionally left blank
■ CONTENTS
Preface
v
1 Introduction: the coming of age of the genome
1
1.1 1.2 1.3 1.4
The age of biology From genetics to genomics A return to function The causes of aging: a random affair
2 12 17 23
2 The logic of aging
27
2.1 2.2 2.3 2.4 2.5
28 36 39 47 52
Aging genes Pleiotropy in aging Interrupting the pathways of aging Longevity-assurance genes Somatic damage and the aging genome
3 Genome structure and function
57
3.1 3.2 3.3 3.4 3.5
58 71 77 81 89
DNA primary structure Higher-order DNA structure Nuclear architecture Transcription regulation Conclusions
4 Genome maintenance 4.1 4.2 4.3 4.4
Why genome maintenance? DNA-damage signaling and cellular responses DNA-repair mechanisms Genome maintenance and aging
91 93 98 105 140
5 Genome instability and accerated aging
151
5.1 5.2 5.3 5.4
152 155 160 177
Premature aging Validity of accelerated-aging phenotypes Genome maintenance and accelerated aging in mice Conclusions
x
CONTENTS
6 The aging genome
181
6.1 6.2 6.3 6.4
183 198 223 229
DNA damage DNA-sequence changes Changes in DNA modification and conformation Summary and conclusions: a DNA damage report of aging
7 From genome to phenome
233
7.1 The causes of cancer 7.2 Genome instability and tissue dysfunction 7.3 Testing the role of genome instability in aging
239 247 278
8 A genomic limit to life?
289
8.1 Aiming for immortality 8.2 SENS, and does it make sense?
289 293
EPILOGUE
299 301 309 353
GLOSSARY REFERENCES INDEX
This page intentionally left blank
Antonie van Leeuwenhoek observes protozoa, bacteria, and germ cells, providing the evidence that life begets life
1677 1735
Carolus Linnaeus publishes the first complete classification of living species
Matthias Schleiden and Theodor Schwann conclude that cells are the basic units of all life forms
1830
Debate between Étienne Geoffroy Saint–Hilaire and Georges Cuvier on form and function
1838 1858
Gregor Mendel presents his basic laws of heredity
1865 1893
August Weismann formulates the first non-adaptive theory of aging
Leslie Orgel proposes the error catastrophe theory of aging
1963 Thomas Kirkwood proposes the disposable soma theory
1984 2003
Aging of the Genome: timeline
Leo Szilard formulates the first somatic mutation theory of aging
1961
1977 Thomas Johnson provides the first evidence for single gene mutations that extend lifespan of an organism
James Watson and Francis Crick propose a double-helical structure for DNA, explaining the perpetuation of genetic information
1956 1958
Peter Mitchell introduces the chemiosmotic hypothesis of energy production
Oswald Avery shows that DNA is the carrier of genetic information
1952 1953
Denham Harman proposes that free radicals are the primary cause of aging
Thomas Hunt Morgan establishes chromosomes as the location of Mendel‘s factors, now termed genes
1937 1944
Peter Medawar formulates the first evolutionary theory of aging
August Weismann recognizes the dichotomy between germ-line and somatic cells
1902 1910
Theodosius Dobzhansky links evolution to genetic mutation
Charles Darwin and Alfred Wallace propose natural-selection theories of evolution
The International Human Genome Sequencing Consortium publishes the complete draft of the human genome sequence
1
Introduction: the coming of age of the genome
Science and technology extend life and improve the quality of life. Whereas in a sense this may have been true since the origin of Homo sapiens, it has never been more apparent than after the Industrial Revolution in the nineteenth century, when great strides in physics, chemistry and medicine significantly improved life for rich and poor alike. By 1900 most European countries had been liberated from the danger of recurrent famine. In addition, improved sanitary conditions, vaccination, and the widespread availability of antibiotics have been responsible for the dramatic increase in average lifespan over the last 200 years. Most of this increase in lifespan has been due to the rapid decrease in infant mortality, since the lives of babies and young children are especially precarious in times of hunger and disease, the latter usually following the former. However, evidence is now emerging that since the 1970s, possibly due to greater awareness of adverse lifestyle habits—such as smoking—and more effective medical care, mortality and morbidity of the elderly has been rapidly declining (at least in developed countries)1,2. In Sweden, a highly developed country with reliable demographic data on human lifespan since 1861, maximum age at death has risen from about 101 years during the 1860s to about 108 years during the 1990s, suggesting that the maximum lifespan of humans and possibly other animals is not immutable3. Whereas average lifespan is deduced from the age at death of all individuals of a population, including those who die very early, maximum lifespan is the maximum attainable duration of life for an individual of a given species. In principle, therefore, the maximum lifespan of our species is the age at death of the longest-lived human, which is 122 years. Jeanne Calment, a French woman who attained this respectable age, died in 1997. A better measure of the trend in achieved human lifespan is the change in upper percentiles of the age distribution of deaths, as was used in the study on maximum lifespan in Sweden cited above. In June 2006 the longest living human was Maria Esther Capovilla from Ecuador, who was then 116 years old. Before her, several human so-called supercentenarians died in quick succession around this age, underscoring the limitations of our species-specific genetic make-up in keeping us alive over extended periods of time. Further optimization in the way we live, even with the best possible medical care, will not appreciably change that situation. Under these ideal conditions, lifespan will likely continue to increase, but slowly and gradually. However, what will happen if science is able to alter the way we are, rather than the way we live? Will the recent dramatic developments in the biological
2
INTRODUCTION
sciences free us from the bonds, which, as in any other species, fix the time of our lives? Is biology crossing a threshold, from a strictly intellectual exercise in understanding life, to an orchestrated effort to halt its demise? Most importantly, can such an effort succeed or are there some inherent mechanistic limitations, which will ultimately prevent us from rapidly achieving, say, a doubling of human lifespan? As I will try to argue in this book, the answers to these questions may be hidden in the genome. The rapid rise of modern biology is very much the story of the coming of age of the genome, the complete set of genetic information of an organism. Genome research has not only provided us with our current basic understanding of the logic of life, but has also supplied the tools to practice a whole new form of biomedicine, now termed genomic medicine. It is the genome as a fluid entity that bears witness to the history of life as it has unfolded on our planet since the first replicators. It is the genome that carries the seeds of our development from fertilized egg into maturity. And it may be the genome, with its inherent instability, that will be responsible for our ultimate demise. In this first chapter I will sketch the major developments in the science of biology, from the Renaissance to the genome revolution, in two parallel lines: one that explains how we gradually gained a mechanistic understanding of how life perpetuates itself through random alterations in DNA, with aging of its carriers as the inevitable by-product, and a much more complicated learning curve that thus far has merely provided the starting points of how we hope to gain a more complete understanding of how life forms are ordered at the molecular level and how this order turns into disorder during aging.
1.1 The age of biology With physics and chemistry at their zenith in the nineteenth and twentieth centuries, biology, the study of life, is often considered the premier science of the century we have just entered, with the promise to revolutionize human existence. The information explosion in biology, which started relatively late, will soon reach a stage when, for the first time in human history, we might be able to extend and improve our life in a more fundamental way than through manipulation of our environment or lifestyle; that is, by intervening in our basic biological circuits in a way that will allow us to break the constraints of our species-specific genetic make-up. To reach this stage, biology has evolved from an originally descriptive science, through a period of hypothesis-driven experimental research, to the data-driven era, which we have now entered, with the prospect of rational interventions based on in silico models that can provide an integrated understanding of the processes that give and maintain human life. At the dawn of modern biology two major, often intertwined, branches of knowledgegathering sprung from the same source: the invention of the microscope in the new
INTRODUCTION
3
permissive era of the Renaissance, which allowed for the first time a detailed observation of the various manifestations of life. A dual quest began to discover life in all its splendid variability and to find out the details of its workings. Along these parallel paths of studying why life is and how it works, the science of aging emerged from the why and how of life’s natural limitation, observed in so many of its individual representatives (see Timeline, p. xi).
1.1.1 THE LOGIC OF LIFE The question of life’s origin and its perpetuation in such a wide variety of forms appeared to be the most challenging of questions and was tackled in successive stages by a number of great minds from the seventeenth to the twentieth centuries. This quest culminated in Darwin’s theory of evolution by natural selection and Watson and Crick’s discovery of the molecular structure of DNA. The grand understanding of the logic of life would prove equally important for understanding its demise: the logic of aging. Before the seventeenth century our state of knowledge was static and, in Western Europe, mainly based on a synthesis of the Greek–Roman heritage and the Christian Church. Following Aristotle (384–322 BC) the general consensus at the beginning of our modern era was that small animals like flies and worms originated spontaneously from putrefying matter. Antonie van Leeuwenhoek (1632–1723) was one of the first to discredit this popular notion of spontaneous generation, based on his direct observations of bacteria, protists, and living sperm cells with home-made microscopes—an early example of technology driving progress in biology. After examining and describing the spermatozoa from mollusks, fish, amphibians, birds, and mammals, he came to the novel conclusion that fertilization occurred when the spermatozoa penetrated the egg. Having reached the consensus that life begets life an explanation was sought for the bewildering variation of life forms on earth. Aristotle had provided the world with a grand biological synthesis, including a classification of animals grouped together in genera and species. He was of the opinion that the current biological diversity had existed from the start, which was later adopted by the church in the form of the dogma that all creatures were created independently of one another by God and organized into a hierarchy. It was Carl Linnaeus (1707–1778) who provided us with a system for naming, ranking, and classifying organisms, still in wide use today, which would become the ultimate tool for recognizing the logic of a system of evolutionary descent. Initially, Linnaeus believed that species weres unchangeable, and he never abandoned the concept of a preordained diversity of life forms. But Linnaeus observed how different plant species could hybridize to create forms which looked like new species. He abandoned the concept that species were fixed and invariable, and suggested that some—perhaps most—species in a genus might have arisen after the creation of the world, through hybridization4.
4
INTRODUCTION
Alfred Wallace (1823–1913) and Charles Darwin (1809–1882), then, provided the now generally accepted explanation for the intriguing similarities among organisms, so beautifully organized by the system of Linnaeus. Whereas the different species had generally been assumed to be immutable and stable since the era of Plato and Aristotle, Darwin had begun to see life as fluid, and recognized that ample variation was present, even among individuals of the same population. Like several scientists before him, Darwin had come to believe that all life on Earth evolved (developed gradually) over millions of years from a few common ancestors. However, the primary mechanism of this process of evolutionary descent was unknown. Based on careful observations of many variations among plants and animals on the Galapagos Islands and South America during a British science expedition around the world, he proposed a process of natural selection to advance certain characteristics best adapted to environmental conditions. The results of this work were published as On the Origin of Species by Means of Natural Selection, or the Preservation of Favoured Races in the Struggle for Life (1859), commonly referred to as The Origin of Species5. Evolution by natural selection was controversial from the beginning and is still less generally accepted than, for example, Einstein’s theories of relativity. This already indicates the sensitivity of society to new concepts in biology involving humans and our position in the living world. The original criticisms of evolutionary descent focused on the need to accept that current life, among which the human species was only one tip on a branching tree, extended back through ancestral species over a time period much longer than the biblical 6000 years. However, the most serious problem, still the main hindrance today for many people to accept Darwin’s theory, is the lack of purpose and direction that speaks from his explanation of life. Natural selection makes use of existing, natural differences among individuals in a population of a species in their suitability to adapt to special problems in their local environment. We now know that such differences in heritable traits continually arise in our germ cells by random changes in the genes that control those traits. Individuals less fit in a given environment are eliminated, whereas those with the most favorable traits leave a disproportionately high number of offspring. As recognized by the great evolutionist Ernst Mayr (1904–2005), the process of adaptation to special problems of local environments gives rise to new species when fragments of a population become geographically and reproductively isolated; this is known as allopatric speciation. (Other, less well explored mechanisms of speciation may also operate.) The concept of open-ended evolution, not necessarily governed by a Divine Plan and with no predetermined goal, is still unaccepted by many. Confusion and resistance to new scientific discoveries are not uncommon, as exemplified by popular reactions to Heisenberg’s uncertainty principle and Freud’s revelations of the subconscious at the beginning of the twentieth century. However, the alarm felt by many when confronted with the implications of Darwin’s theory regarding the position of humans in life as a
INTRODUCTION
5
whole are quite unique. Indeed, the validity of the physical principles underlying the automobile, air travel, and the personal computer are never doubted by the general public. By contrast, equally solid principles in biology are often rejected out of hand by sizable segments of the educated public based on the strong intuitive appeal— often inspired by religion—of intelligent design and purposeful direction. Biology will continue to raise feelings of uneasiness in the years to come. After Darwin, the next major development in biology was the emergence of the concept of the gene. A problem with Darwin’s theory of natural selection as the mechanism of evolutionary change was the lack of knowledge as to how random variations in heritable traits could arise and how they could be perpetuated from parents to offspring. Ironically, the genetic principles governing this latter process had already been described in Darwin’s lifetime by the Czech monk, Gregor Mendel (1822–1884). Working with different kinds of peas, Mendel demonstrated that the appearance of different hereditary traits followed specific laws, which could be understood by counting the diverse kinds of offspring produced from particular sets of parents. He established two principles of heredity that are now known as the law of segregation and the law of independent assortment, thereby proving the existence of paired elementary units of heredity (which he called factors) and establishing the statistical laws governing them. Mendel’s findings on plant hybridization were ignored until they were confirmed independently in 1900 by three botanists. After 1900, the physical basis for Mendel’s laws was discovered in the form of the chromosomal basis for the transmission of genes from parents to offspring. Thomas Hunt Morgan (1866–1945) was the first to provide conclusive evidence that chromosomes are the location of Mendel’s factors, termed genes by Wilhelm Johanssen in 1907 (in Greek meaning ‘to give birth to’). Morgan chose the fruit fly, Drosophila melanogaster, as his experimental animal, which has remained a key experimental model system in genetics ever since. In 1910, he found a mutant male fly with white rather than the normal red eyes. Since all the female flies had red eyes with only some males having white eyes, Morgan realized that white eye color is not only a recessive trait but is also linked in some way to sex. This work led to the identification of four so-called linkage groups, which correlated nicely with the four pairs of chromosomes that Drosophila was known to possess. Their subsequent breeding experiments provided proof that the chromosomes are indeed the bearers of the genes, with different genes having specific locations along specific chromosomes. Traits on one particular chromosome naturally tended to segregate together. However, Morgan noted that these ‘linked’ traits would separate, from which he inferred the process of chromosome recombination: two paired chromosomes could exchange genetic material between each other, an event termed crossover. The frequency of recombination appeared to be a function of the distance between genes on the chromosome. The smaller that distance, the greater their chance of being inherited together, whereas the farther away they are from each other, the more chance of their
6
INTRODUCTION
being separated by the process of crossing over. The Morgan is now the unit of measurement of distances along all chromosomes in fly, mouse, and human. In the meantime, cytologists had described the processes of mitosis and meiosis at the end of the nineteenth century. The chromosomes, thread-shaped structures under the microscope, were known to be located in the nucleus of a cell, but nobody knew their function. By correlating their breeding results with cytological observations of chromosomes, Morgan’s group provided the physical reality for Mendel’s hypothetical factors. It was recognized that chromosomes, which could be distinguished, quantified, and observed to occur in pairs, except in germ cells, housed the genetic material. Germ cells were demonstrated to have only one copy of each chromosome pair, with fusion of the germ-cell nuclei restoring a complete set of chromosomes, half from the father and half from the mother. A late highlight in this development was the work of Cyril Darlington (1903–1981), who made the connection between the structural behavior of chromosomes, including the mechanics of chromosomal recombination, and the functional consequences in terms of heredity6. The chromosomal theory of inheritance, with its distinction between somatic and germ cells, ended speculation by Darwin, Jean-Baptiste Lamarck (1744–1829), and others that offspring were a mere blending of the parents and that acquired traits could be inherited. It was also around this time that the terms phenotype and genotype began to be distinguished. The phenotype of an individual organism comprises its observable traits (such as size or eye color) whereas the genotype is the genetic endowment underlying the phenotype. Of note, in those early days the genotype could only be determined on the basis of the phenotype because the nature of the genetic material was still unknown. Therefore, inheritance patterns could only be checked by breeding experiments. Based on the early separation between somatic and germ cells, August Weismann (1834–1914) first formulated the unidirectional theory that the phenotype cannot affect the genotype7. The distinction of germ line and soma would profoundly influence our ideas about aging. Weismann recognized that the germ cells are not affected by any variation that might occur in an individual. This is especially relevant for somatic changes in the structure of deoxyribonucleic acid (DNA), which we now know is the carrier of the genetic information. Such changes, termed mutations, in a somatic cell may damage the cell, kill it, or turn it into a cancer cell. But, whatever its effect, a somatic mutation is doomed to disappear when the cell in which it occurred or its owner dies. By contrast, germ-line mutations such as the one that gave rise to Morgan’s white-eye trait, will be found in every cell descended from the zygote to which that mutant gamete contributed. If an adult is successfully produced, every one of its cells will contain the mutation. Included among these will be the next generation of gametes, so if the owner is able to become a parent, that mutation will pass down to yet another generation. Mutations in somatic cells may be expressed, but are not passed on to further generations. Mutations in germ cells can be both expressed and transmitted to descendents.
INTRODUCTION
7
The distinction between germ line and soma exists only in animals. In plants, cells destined to become gametes can arise from somatic tissues. In organisms without sexual reproduction, such as many unicellular organisms, there is no distinction between germ and soma. In Weismann’s view, the soma simply provides the housing for the germ line, seeing to it that the germ cells are protected, nourished, and combined with the germ cells of the opposite sex to create the next generation. This provided the logical basis for rejecting the ideas of Lamarck and others that characters acquired during lifetime could be inherited by the next generation. Weismann’s views foreshadowed the concept by Richard Dawkins (Oxford, UK) of the gene as the fundamental unit of selection, instead of species, group, or individual8, as well as the disposable soma theory of aging by Tom Kirkwood (Newcastle upon Tyne, UK)9. Weismann was also the first to explain the aging of metazoa in evolutionary terms. In the first instance he proposed that aging was an evolutionary adaptation to avoid the need for offspring to compete with their parents for scarce resources. The idea that old individuals die as an act of altruism to the rest of the group or species is now generally considered as naive and incompatible with the negligible impact of aging on animals in the wild (few animals survive long enough to experience old age). However, Weismann also presented the case for aging as a non-adaptive trait, which would again foreshadow modern thinking about why we age. In this case, he argued that characters that have become useless to an organism, such as eyesight in animals that never see the light, are not subject to natural selection. Applied to the ‘useless period of life following the completion of reproductive duty’ this theory would predict a weakening of selection against characters with adverse effects later in life. Moreover, it predicts the positive selection of such traits if there is some benefit in the earlier years of life7. In the 1940s, Weismann’s neodarwinism was integrated with new findings in laboratory genetics and fieldwork on animal populations. This so-called evolutionary synthesis, in a sense the grand finale of the work begun by Darwin and his predecessors, started with T.H. Morgan, mentioned above, and reached a new height during the first decades of the twentieth century with the work of the great mathematical population geneticists, Ronald Fisher (1890–1962), Sewall Wright (1889–1988), and J.B.S. Haldane (1892–1964). They developed quantitative genetics as a synthesis of statistics, Mendelian principles, and evolutionary biology. They demonstrated that the same principles that applied to discrete traits (such as eye color) were also valid for quantitative traits, such as height and certain behavioral characteristics, which display continuous variation in the population. These concepts were later combined with explanations for the origin of biodiversity by Theodosius Dobzhansky (1900–1975), the previously mentioned Ernst Mayr, and others, resulting in the integration of Mendel’s theory of heredity with Darwin’s theory of evolution and natural selection. The unification of genetics and evolution by natural selection also gave rise to the first discussions—in the new, mathematical language of the modern synthesis—of the
8
INTRODUCTION
evolutionary basis of aging. It was Fisher who noticed, probably for the first time, that the chance of individuals to contribute to the future ancestry of their population declines with age10. Later, this would lead Peter Medawar (1915–1987), a Nobel laureate and better known for his work on transplantation immunology, to propose that aging, at least in sexually reproducing organisms with a difference between the soma and the germ line, is a result of the declining force of natural selection with age (see Chapter 2 in this volume). What was still not clear at the time was the nature of a gene and the mechanism of Mendel’s transmission of heritable traits through the germ line. It was only in 1944 that Oswald Avery (1877–1955) and collaborators made a convincing case for DNA as the carrier of the genes11. They were studying a substance that could turn non-pathogenic variants (R cells) of Streptococcus pneumoniae, a bacterium that causes pneumonia, into pathogenic ones (S cells). This so-called transforming principle, which had a high molecular mass, was resistant to heat or enzymes that destroy proteins and lipids, and it could be precipitated by ethanol. Hence, it was most likely DNA, a substance already described by Johann Friedrich Miescher (1844–1895) in 1869 as occurring in human white blood cells and in the sperm of trout. However, the nature of the genetic code and a mechanism for how DNA was able to transfer this information from cell to cell and how it could convert this information into cellular function was still unknown. James Watson (Cold Spring Harbor Laboratory, NY, USA) and Francis Crick (1916–2004) provided the answer in 1953 in the form of the molecular structure of DNA: two helical strands of alternating sugar-phosphate sequences, each coiled round the same axis, held together by adenine–thymine- and cytosine–guanine-specific base pairing. The base pairing properties of DNA dictate the mechanism of gene replication12. Hence, it was now known that the complete set of genetic information of an organism, the genome, was written in its DNA. Genomes, which can vary widely in size, from 600 000 bp in a small bacterium to 3 billion in a mammal, were subsequently demonstrated to be the repository of the genes, the basic physical and functional units of heredity. The years immediately after Watson and Crick are now known as the classical period of molecular biology. First, Matthew Meselson (Cambridge, MA, USA) and Franklin Stahl (Eugene, OR, USA) experimentally confirmed13 the process of semiconservative DNA replication predicted by the double-helical, base-paired model proposed by Watson and Crick. DNA isolated from Escherichia coli after growth in medium containing heavy or light isotopes of nitrogen showed a distinct density distribution in CsCl gradients. After switching medium, DNA of an intermediate density was obtained, which is expected if the newly replicated DNA is a hybrid molecule consisting of one parental and one newly synthesized strand. Then, following the prediction by François Jacob (Paris, France) and Jacques Monod (1910–1976) that messenger ribonucleic acid (mRNA) transcribed from the DNA of a gene in the form of a single-strand complementary copy was the template for protein synthesis14, Crick, Sydney Brenner (San Diego, CA, USA) and colleagues15 demonstrated in 1961, by deleting bases one by one from DNA of the bacteriophage T4, that the genetic
INTRODUCTION
9
code was a triplet of bases. A string of triplets specifies the full sequence of amino acids in a protein chain. Using a cell-free translation system and synthetic homopolymers, Marshall Nirenberg (Bethesda, MD, USA)16 and Har Gobind Khorana (Cambridge, MA, USA)17 identified which codons corresponded to which amino acids. Meanwhile, the laboratories of Mahlon Hoagland (Worcestor, MA, USA)18, Robert Holley (1922–1993)19, and others had discovered transfer RNA (tRNA), predicted by Crick in his adaptor hypothesis as the entity that recognized triplets of bases on the mRNA. Adaptor enzymes link each kind of amino acid to the appropriate carrier, tRNA. Protein synthesis or translation is carried out by bringing the mRNA and the set of tRNAs charged with the appropriate amino acids to the ribosomes, discovered earlier as the protein-making apparatus in the cytoplasm. The guiding role of Francis Crick in bringing this classical period to its zenith is now well recognized. Crick’s predictions that the genetic code was universal to all forms of life and that genetic information can go only one way—that is, from DNA via RNA to protein—proved correct with minor exceptions. This so-called central dogma of molecular biology is another way of saying that acquired characteristics cannot be inherited. With the discovery of the structure of DNA and the genetic code, the origin of Darwin’s existing natural differences in heritable traits had also become clear. DNA in the living cell is not completely stable, but can undergo alterations in its base pair composition through errors during replication or the repair of chemical damage. Hermann Joseph Muller (1890–1967), a student of T.H. Morgan, had already demonstrated in 192720 that mutations could be induced by radiation. He identified mutations mainly by the observed effect on an organism, but was able to show that mutations can result from breakages in chromosomes and changes in individual genes. He also realized that the majority of such random mutational changes are deleterious, although an occasional mutation is beneficial, for example, by giving rise to a better-functioning protein. However, as we now know, the genetic code is tolerant of certain mutations. This degeneration of the code is due to the fact that there are three times as many codons as there are amino acids, hence the tolerance of some amino acids for a mismatch at the third position of each triplet. With the evolutionary synthesis and the new understanding of its underlying molecular principles, the pursuit of the origin and perpetuation of life was essentially over. From now on, biology could fully focus on unraveling the structure and function of life’s various manifestations.
1.1.2 SEARCHING FOR STRUCTURE AND FUNCTION The desire to know all structural, organizational, and functional facets of life sprung from the same source as the theory of evolution and modern genetics: the careful observations made by the pioneers of science in the seventeenth century, with microscopy as their
10
INTRODUCTION
main tool. Then, as now, there was a significant relationship between the ability of craftsmen to provide good instrumentation and the direction of scientific investigation. Most notable among these early scientists, apart from the above-mentioned Antonie van Leeuwenhoek, was his contemporary, Marcello Malpighi (1628–1694). Malpighi was probably the first scientist to use model organisms—frogs and turtles—to obtain structural information on human organs, thereby inventing comparative anatomy. Following the early work of William Harvey (1578–1657) on human blood circulation, Malpighi discovered blood flow through capillaries in the lungs, opening the way to understanding the function of this organ in respiration. He conducted a famous comparative study of the liver, from snails through fishes and reptiles, right up to humans, and he was the first to give an adequate description of the development of the chick in the egg21. At this time, it had begun to dawn from Leeuwenhoek’s work, as well as from microscopic observations by the great British natural philosopher Robert Hooke (1635–1703), that life was organized around a basic unit, termed a cell by Hooke. However, it took until 1839 before Mathias Schleiden (1804–1881) and Theodor Schwann (1810–1882) could make the conclusion that cells were the basic units of life. In animals, cells were progressively organized into tissues, organs, systems, and, finally, the whole body. The adult human body is an aggregate of more than 75 trillion cells. With the birth of modern cell theory, anatomists had widened their scope and new disciplines emerged, such as embryology, cytology, and physiology, all focused on understanding the mechanisms of life in all its facets, and how this unfolds from a fertilized egg to an adult organism. Meanwhile, in studying various life forms, the early scientific community was struggling with the question of whether organisms were integrated wholes, as advocated by Georges Cuvier (1769–1832), or whether morphology could be changed and affected by environmental conditions, as proposed by Étienne Geoffroy Saint-Hilaire (1772–1844). In other words, does function strictly dictate form with no modification possible, or do body plans constrain how organ functions are manifested? These positions, which were later synthesized, remain a leitmotiv for modern systems biology and functional genomics. The dramatic increase in our understanding of how structure follows function was a result of the application of new insights in chemistry, most notably organic chemistry, to study different cellular components. This would first lead to biochemistry, the science dealing with the chemistry of living matter, and ultimately to molecular biology, the branch of biology dealing with the nature of biological phenomena at the molecular level through the study of DNA, RNA, proteins, and other macromolecules involved in genetic information and cell function. The undisputed highlight of this development was our ultimate understanding of how cells harvest the energy of food through the conversion of adenosine diphosphate (ADP) into the energy-carrying compound adenosine triphosphate (ATP) in subcellular structures called mitochondria. In his 1961 paper22, Peter Mitchell (1920–1992) introduced the chemiosmotic hypothesis, connecting the electrontransport chain, through a proton (H+) gradient across the inner mitochondrial membrane,
INTRODUCTION
11
with oxidative phosphorylation and the synthesis of ATP. Critically important to all biology and shaping our understanding of the fundamental mechanisms of this most important of all cellular activities, the elegance of the chemiosmotic model in correlating structure and function would have been appreciated by Cuvier. The universality of the process of oxidative phosphorylation suggests its importance as a factor in aging. Ironically, even before Mitchell’s landmark paper, another chemist, Denham Harman (Omaha, NE, USA), proposed that free radicals, the adverse by-products of oxidative phosphorylation, were a ubiquitous cause of aging23. This hypothesis is known as the free radical theory of aging and has been with us ever since. Free radicals are now generally considered as a most likely explanation for the damage that ultimately leads to our demise. It also drew attention to the mitochondria and their own independent genome, so close to the origin and main source of free radicals. Distinct from the far larger nuclear genome, the mitochondrial genome is now considered a major target for spontaneous mutagenesis. In turn, this may adversely affect the process of oxidative phosphorylation itself, thereby accelerating formation of free radicals. This is described in detail in Chapter 6. As we have seen, molecular biology provided the insight that proteins were the workhorses of biological systems, and DNA the carrier of genetic information, organized in the form of a genome. Genes were shown to be specific sequences of base pairs that contain the instructions, in the form of a triplet code, for making proteins. Interestingly, not long after the discovery of the fundamental mechanism of protein biosynthesis, Leslie E. Orgel (San Diego CA, USA) proposed in 1963 that cellular aging involves the accumulation of defective proteins as a result of an inherent inaccuracy of the translational machinery. This is generally known as the error catastrophe theory of aging and longevity, based on Orgel’s realization that the faulty RNA and DNA polymerases, also resulting from translational errors, could lead to an exponential increase of defects in protein, RNA, and DNA, causing the collapse of the cellular machinery for information transfer. This idea is not supported by experimental evidence, but it can be argued that errors are random, with each cell acquiring a unique set of errors. Since current technology is geared towards analyzing mixtures of cells rather than individual cells, we may simply be unable to detect error catastrophes. In the decades following the discovery of the double helix, and especially after the development of recombinant DNA technology, molecular biology became a premier discipline in biology, always at the cutting edge of new developments. Initially, molecular biology remained separate from more traditional disciplines, such as physiology. However, gradually these other disciplines would include molecular biology as an aide in support of their own research endeavors. Meanwhile, the realization of the extreme complexity of the gene–phenotype relationship necessitated a whole new approach, which coincided with the informatics explosion, bringing powerful new computers and the internet. Eventually this would lead to a departure from the original reductionist
12
INTRODUCTION
approaches to holistic strategies, providing a more comprehensive understanding of life, and the emergence of functional genomics and systems biology.
1.2 From genetics to genomics In the heydays of molecular biology it seemed natural to begin our effort of understanding the structure and function of various life forms with understanding individual genes and their activities in different organisms. Indeed, after Watson and Crick, the central dogma may have clarified the mechanisms underlying Mendel’s laws, but virtually all known genes were still identified only by mutations and their phenotypic consequences. Genetics was a matter of studying inherited phenotypes, rather than genes, none of which had been isolated before 1973, when Stanley N. Cohen of Stanford University and Herbert W. Boyer of the University of California, San Francisco, developed the laboratory process to take DNA from one organism and propagate it in a bacterium. This process, called recombinant DNA technology, was used in 1977 for the production of the first human protein manufactured in a bacterium: somatostatin, a human growth hormonereleasing inhibitory factor. For the first time, a synthetic, recombinant gene was cloned and used to produce a protein24. The following decade saw a surge in the study of genes and their function, for which Tom Roderick (Bar Harbor, ME, USA) in 1986 coined the term genomics. Genomics was highly technology-driven, as exemplified by the rapid emergence of a host of new techniques and instruments. The undisputed highlight of this development was the discovery, by Kari Mullis, then at the Cetus Corporation, of the polymerase chain reaction (PCR), a technique for amplifying DNA sequences in vitro by separating the DNA into two strands and incubating them with oligonucleotide primers and DNA polymerase (Fig. 1.1). PCR can amplify a specific DNA sequence as many as one billion times, and quickly became essential in biotechnology, forensics, medicine, and genetic research as probably no method before. Initially, genomics was not different from standard, investigator-initiated research and was entirely hypothesis-driven. This would change with the conception of the Human Genome Project (HGP), the international research effort that determined the DNA sequence of the entire human genome. The rationale behind the HGP was that by sequencing a complex genome, the amino acid sequences of all proteins as well as all sequenceencoded regulatory and structural characteristics of that genome would be immediately available, obviating the need to purify and characterize each feature separately. Cloning genes into expression vectors allowed the production of proteins, but also allowed their engineering, for example, for studying their phenotypic characteristics in cell cultures or experimental animals. Indeed, it was around this time—in the 1980s—that the methods to make transgenic mice were developed by Jon Gordon (New York, NY, USA)25,
INTRODUCTION
13
Targeted sequence 39
59
59
39
Primers, DNA polymerase, dNTPs
Heat, cool, and 59 DNA synthesis
39
59
39 59
First cycle 59 39
59 Repeat
Second Cycle
Repeat After 25 cycles, the target has amplified 106-fold
Fig. 1.1 In the PCR small single-stranded DNA fragments, complementary to known sequences that flank a nucleic acid sequence, are used as primers (black rectangles) to amplify this sequence millions of times through 25 or more cycles of in vitro enzymatic synthesis. dNTPs are the four deoxynucleotide Triphosphates.
Ralph Brinster (Philadelphia, PA, USA) and Richard Palmiter (Seattle, WA, USA)26, and their co-workers. The first use of transgenic mice was to study gene function in the whole animal, in particular how and why a specific gene is turned on in some tissues and turned off in others. This diversity of gene expression that produces the distinct cell types and tissues of the body, making a muscle cell different from a liver cell, had quickly become of central interest in molecular biology. Access to comprehensive genome sequences of different species allowed scientists to systematically address this question.
14
INTRODUCTION Region to be sequenced 3'
5'
5'
3' Heat, add primer
5'
3' 5'
3'
5' 3'
3' 5'
dd TT
T A
5' 3' T
TP
P
A
ddG
CT
5'
dd
TP
ddA
3'
DNA polymerase P
dATP, dTTP, dCTP, dGTP
3' 5'
C C
3' 5'
5' 3'
G
5' 3'
G T T C G G
Electrophoresis
G A T T A C
Fig. 1.2 Principle of nucleotide sequencing by the Sanger method. In this case a single primer (black rectangle) is used to generate a set of fragments with a common 5’ origin through basespecific interruption of in vitro enzymatic synthesis. A, adenosine; C, cytidine; G, guanosine; T, thymidine; dATP, dCTP, dGTP, dTTP, deoxy-adenosine, -cytidine, -guanosine, and -thymidine triphosphate, respectively. ddATP, ddCTP, ddGTP, ddTTP, dideoxy-adenosine, -cytidine, -guanosine, and -thymidine triphosphate respectively.
It was realized early on that the average research laboratory was too small to contribute significantly to such a project and that methods of scale were needed. This resulted in most of the work being done by large genome centers. Contributors to the HGP included the US National Institutes of Health and the US Department of Energy (where discussions
INTRODUCTION
15
of the HGP began as early as 198427), numerous universities throughout the USA, and international partners in the UK, France, Germany, Japan, and China. A separate, commercial project to sequence the human and other genomes was initiated in 1998 by the Celera Genomics Corporation28. In the course of completing the sequence, two separate interim working drafts of the human genome were produced in 2001 with much publicity by both the public consortium29 and Celera30. However, the major aim of the HGP was to obtain a comprehensive, ‘finished’ sequence of the entire 3109-base haploid human genome, which was eventually published in 2004 and contained 2.85 billion nucleotides, covering 99% of the euchromatic genome. The paper in Nature reporting this accomplishment had over 2800 authors, an example of the emergence of large consortia of researchers at the expense of the more classical investigator-initiated approach of the past31. To the surprise of many, the complete sequence of the human genome did not immediately tell us how many human genes there are. This uncertainty is likely to last for a while because of the limitations in gene-prediction software, which thus far have precluded an accurate assessment of the total number of human protein-coding genes. It is generally thought that this number is somewhere around 30 000. In hindsight, the achievement of its primary goal may have been less important than the impact of the HGP on the way biological research was conducted. Its legacy will probably always be associated with the transformation of biology from an almost exclusively solitary, hypothesis-driven science into an information science. This was based on the use of high-throughput methods and the increasing need to organize research as large collaborative efforts of multiple investigators from various disciplines. However, rather than abandoning individual, hypothesis-driven research, this development is more likely to eventually lead to the iterative and integrative approach of global analyses driven by hypothetical models, now known as a systems approach. I will discuss this extensively below. The main driving force behind the globalization of biological research was the additional goal of the HGP to develop novel technologies and improve existing ones. The success of the HGP in converting regular methods in molecular biology into methods of scale is exemplified by the great improvements in the sequencing method first described by Frederick Sanger (Cambridge, UK) and co-workers in 197732. This method is based on the use of a DNA polymerase to extend, in four separate reactions, an oligonucleotide primer from its annealing site at the beginning of the target sequence over a length of 500–1000 bp. Each reaction contains all four deoxynucleotide triphosphates plus a limiting amount of either adenine, thymine, cytosine, or guanine dideoxynucleotide triphosphate, which terminate the reaction upon incorporation. After electrophoretic separation of each fragment mixture at a resolution of 1 bp, the sequence of the target can be read directly from the resulting banding pattern (Fig. 1.2). Initially, radioactive labeling was used to detect the fragments after size separation. Later, fluorescent labels were developed, which allowed automated detection of the electrophoretically separated fragments using a laser33. From this first partial automation
16
INTRODUCTION
of the Sanger dideoxy sequencing principle to the current, almost fully automated 384-capillary electrophoresis systems there has been an approximately 2000-fold increase in throughput. Ironically, this has not been achieved by new technology, as originally anticipated, but almost exclusively by the introduction of methods of scale. Although such improvements have now made it relatively easy to obtain the consensus sequence of a genome quickly, especially that of a small microbe, conventional sequencing is still not cost-effective enough for routine application, for example, in large-scale genetic epidemiology or clinical diagnosis. It is anticipated that novel sequencing principles, including single-molecule sequencing34, will successfully address remaining limitations in cost, speed, and sensitivity. In this respect, it has been predicted that about 10 years from now it will be possible to sequence an entire human genome in 30 min for about $100035. In addition to the human genome, hundreds of genomes of other species, from simple microorganisms, such as E. coli36, to the mouse, rat, and chimpanzee37, have now been sequenced completely. The information that can be derived from these sequences is vast. Large-scale sequencing totally transformed certain disciplines as genetics and physiology and created new ones, such as comparative genomics, the most powerful way to elucidate the roles of many related genes. Although the practice of comparing gene or protein sequences with each other, in the hope of elucidating functional and evolutionary significance, is well established, its application to complete genomes greatly expands its utility and implications. For example, phylogenetic trees can be built not from the sequences of a single gene (usually ribosomal RNA (rRNA) genes) but from multiple gene sequences as well as from non-sequence information, such as similarities in gene repertoire and gene order38. This requires the rational classification of genes and proteins, which is usually done in the form of a system of orthologous gene sets. (Orthologs are homologous genes that evolved from a single ancestral gene in the last common ancestor of the compared genomes; paralogs are genes related via duplication within a genome.) Major applications of cross-species genome comparisons are the identification of functionally important genomic elements, e.g. protein-coding and regulatory sequences, on the basis of homology39. This is based on the assumption that functionally important regions tend to have a lower mutation rate than non-functional regions. The rapid increase in wholegenome sequences from different mammals and the development of better tools for their comparison should lead to increased insight into the functional constraints of the human genome. The HGP has also been the starting point for several new, large-scale initiatives in genomics. For example, the establishment of a catalog of all common sequence variants (single nucleotide polymorphisms or SNPs) in the human genome with their patterns of linkage disequilibrium (the HapMap project) has been initiated to facilitate the identification of genetic risk factors in disease susceptibility and other phenotypes, whereas the Encyclopedia of DNA Elements (ENCODE) project aims to identify different functional
INTRODUCTION
17
elements in the human genome40. The latter is very much based on the realization that gene function can only be understood in the context of the genome as a whole, with its multiple overlapping networks of regulatory sequences. All these large-scale genome projects are part of a general development to collect biological information in a systematic way and make it publicly available. The rapid growth of the Internet around the same time greatly facilitated the use of such shared resources, which now play a crucial role in conducting biological research and have become the basis for functional genomics and systems biology.
1.3 A return to function Complete DNA sequence information is not the end, but merely the beginning of our quest to understand how genomes—and therefore organisms—function and how time, both evolutionary time and the lifetime of an individual, can affect such function. For this purpose, genes need to be identified; the function of their products (RNAs and proteins) must be elucidated and the role of non-coding regulatory sequences needs to be understood. Since the landmark completion of the HGP, the type of biology focused on the identification and functional analysis of genes, coding regions, and other functional elements of entire genomes on a high-throughput basis has been termed functional genomics. Whereas genomics implies the study of genes and their function, functional genomics attempts to integrate all genes, their products, and their resultant phenotypes into dynamic networks of molecular pathways that ultimately determine our physiology (Fig. 1.3). Such networks of interaction have now all but replaced the original onegene/one-protein way of thinking. If the advances of molecular biology and genomics had made anything clear, it was the stupendous complexity of living cells and their interactions to generate complete functioning organisms. A discrete biological function can only rarely be attributed to one individual protein encoded by one gene. In reality, biological characteristics involve complex interactions among many components in cells, such as DNA, RNAs, proteins, and small molecules. Emphasizing this integrative nature of biological function, Hartwell and Hopfield coined the term functional modules41. In functional genomics, a distinction is made between the different levels of organization in the cell. The genome, as we have seen, denotes the totality of all genes on all chromosomes in the nucleus of a cell. The complete set of mRNAs, the next hierarchical level below the genome, is called the transcriptome. Next, there is the proteome, which is the set of all expressed proteins for a given organism. This is followed by the metabolome, a biochemical snapshot of the small molecules produced during cellular metabolism, such as glucose, cholesterol, and ATP, and
18
INTRODUCTION
HIF1A
CDK4
INK4A
MDM2
TNF
AKT
PTEN
ILK
TNFR p21
BCL2
RB1
p53 RBBP8
BRCA1
LMO4
NFB
ERBB2
BAX
CCND1
PI3K
p73
Fig. 1.3 An example of a network showing interactions between the TP53 protein and other gene products. Solid lines indicate protein–protein interactions whereas dashed lines show activation or repression at the transcriptional level.
several other comprehensive sets of biological information, such as the secretome (total of secreted molecules) and the interactome (a complete set of macromolecular interactions, such as protein–protein interactions). Functional genomics was driven by the need to understand the formal relationships between genes and all the -omes, including the rules that control transition between these levels and from them to complex, functional systems, such as oxidative phosphorylation, genome maintenance, and the immune system. A key aim now became to systematically catalog all molecules and their interactions in a living cell. To do this, high-throughput methods for genome-wide data collection have become indispensable. Since the discovery of genes has outpaced our capacity to understand their biology, high-throughput methods to assess genotype–phenotype relationships are rapidly being developed and applied. The most popular vehicle for high-throughput analysis in biology has become the microarray chip. In its first successful manifestation, hundreds to thousands or tens of
INTRODUCTION
19
thousands of cDNAs or oligonucleotides, complementary to parts of individual mRNAs, were attached to a glass slide. Hybridization of such slides with labeled probes obtained from reverse-transcribed RNA from tissues or cells of interest permits the analysis of changes in expression of a large number of genes simultaneously. This technology has the ability to reveal patterns of gene expression across different samples. For this purpose, genes are grouped into classes with similar profiles of activity, in an approach called cluster analysis. Such genes may have related functions or be regulated by common mechanisms. The structured gene-functional-categorization database, Gene Ontology or GO, provides the opportunity to partition genes into functional classes. Microarrays are now also used to study DNA sequence variations (SNPs), proteins, protein–protein and protein–DNA interactions, and various other structural characteristics of the cell or tissue. Another tool that can be applied in a microarray format on a genome-wide scale, RNAmediated interference (RNAi), proved to be of critical importance in bridging the gap between genotype and phenotype that had opened up since T.H. Morgan. First demonstrated in 1998 in Caenorhabditis elegans42, RNAi allows the sequence-specific silencing of genes using synthetic double-stranded RNAs. Such exogenous RNAs co-opt a ubiquitously expressed, evolutionarily conserved gene-regulatory system consisting of microRNAs (miRNAs; Chapter 3). Endogenous miRNAs are transcribed as single-stranded precursors up to 2000 bp in length and exhibit significant secondary structure, resulting in stems and loops. Such primary transcripts are first processed in the nucleus and after entering the cytoplasm converted by the RNase III enzyme Dicer into double-stranded 21–23-nucleotide-long mature RNAs. Synthetic forms of miRNAs, so-called short hairpin RNAs (shRNAs), can be used experimentally to mimic their natural equivalents. Alternatively, it is possible to use short interfering RNAs (siRNAs), synthetic doublestrand RNAs of less than 30 bp. Such siRNAs bypass cleavage by Dicer. All the small double-stranded RNAs need to associate with the RNA-induced silencing complex (RISC), which unwinds them and associates stably with the strand that is complementary to the target mRNA. Depending on the degree of homology, the complexes inhibit gene activity either by translational repression or triggering mRNA degradation. Apart from using synthetic variants of these RNAs, it is also possible to express them using a plasmid to silence gene expression for longer periods of time43. RNAi is a typical example of reverse genetic technology and can conveniently be applied in a microarray format. RNAi at such a genome-wide scale was applied early on in the science of aging to screen for genes regulating lifespan (Chapter 2). In general, variation greatly increases from DNA to RNA to protein. For example, while the entire human genome may contain no more than 30 000 genes, there may be three times that many proteins, due to alternative splicing; this is the production of more than one transcript by including or excluding specific exons (the DNA segments of a gene that are protein-coding; see Chapter 3) or altering the length of a specific exon. This is without
20
INTRODUCTION
taking into account posttranslational modification, such as the attachment of phosphate, acetate, lipid, or sugar groups. To systematically describe the proteome and its different patterns of interaction in a complex organism is therefore more difficult than making an inventory of all genes. This is especially true because microarray technology for proteins is less well developed as it is for genes. Nevertheless, progress in this field is now also rapid, resulting in ever larger sets of proteins often subdivided according to their specific modification. For example, protein phosphorylation is estimated to affect 30% of the proteome and is a major regulatory mechanism that controls many basic cellular processes. Using microarray technology, the in vitro substrates recognized by most yeast protein kinases were recently identified, involving over 4000 phosphorylation events and 1325 different proteins. This collection of data was called the phosphorylome44. With phenotypic variation much more extensive than genotypic variation and an increasing number of global data-sets emerging at ever-shorter time intervals, the resulting deluge of data is truly transforming molecular biology, from the focused analysis of single genes and proteins to the systematic analysis of entire networks of coupled biochemical reactions and feedback signals. In this respect, the HGP has taught us to see the study of genomes as information science that requires support by advanced computational biology tools and databases. The new discipline of bioinformatics plays a critical role in implementing this endeavor. Bioinformatics uses information technology to organize, visualize, interpret, and distribute biological information to answer complex biological questions. It allows workers in functional genomics to cope with the flood of data and address biological questions in a fraction of the time it would take using traditional analysis techniques. It is bioinformatics that enables functional genomics to bring order out of a vast number of data points. A central component of functional genomics, driven by high-throughput methods and information science, is the ability to standardize extensive sets of disparate data. A key attribute in standardizing biological databases, which makes them computationally accessible, is an ontology. An ontology formally defines a common set of terms that are used to describe and represent a domain. Such vocabularies of terms specify the concepts in a given field and avoid semantic confusion. An example is the Gene Ontology, which can be used to describe the biological process, molecular function, and cellular location of any gene product. Ontologies have also become important in systematically collecting phenotypic information. As mentioned above, the term phenotype refers to observable traits and can be applied to any morphologic, biochemical, physiologic, or behavioral characteristic of an organism. The complete phenotypic representation of a species is now known as its phenome. The Mouse Phenome Project is a consortium of academic and industrial participants that promotes the quantitative phenotypic characterization of a defined set of mouse strains under standardized conditions. Such coordinated efforts in obtaining phenomic databases are now replacing phenotypic investigations carried out by thousands of independent investigators throughout the world, most of whom have no
INTRODUCTION
21
communication with each other. Standardized and comprehensive, such databases are critically important in the unraveling of molecular networks in the context of a functional unit, such as an organ or an organism. An important condition in using large, standardized data-sets in an efficient manner for testing hypotheses and generating new ones is their integration. A major challenge in bioinformatics is to create single platforms for the integrated analysis of multiple, distributed data sources; for example phenotypic data with protein-interaction data and geneexpression data. In other words, the different components of the biological landscape in the form of the different -omics levels are pulled together to gain an understanding of biology at a higher level. This convergence will happen in an approach known as systems biology (Fig. 1.4). Systems biology studies the interrelationships of all the elements in a
Systems biology
Predictive modeling Phenome
Computational analysis
Metabolome
Interactome
Functional genomics
High-throughput measurements
Proteome
Transcriptome Perturbation
Genome
Fig. 1.4 The holistic science of systems biology attempts to define biological realities on the basis of global responses of cells, organs, or entire organisms to environmental or genetic perturbations.
22
INTRODUCTION
system rather than studying them one at a time, in an effort to uncover hidden rules governing the ensemble of biomolecules working concertedly to perform certain functions in the cell. It aspires to use comprehensive data-sets, including such specifics as the experimental conditions under which the data were obtained, for building predictive models. In a sense, systems biology is the antithesis of the reductionist approach to biology, which has been so successful in the past in providing insights into the molecular machinery of many living systems and will continue to do so in further unraveling gene function in functional genomics approaches. However, it did not and cannot provide an understanding as to how molecular processes are integrated to provide function and how molecular function is regulated in living cells so as to give rise to dynamic cell, tissue, and organismal phenotypes. General systems theory was conceived in the 1930s by the Austrian biologist Ludwig Von Bertalanffy (1901–1972), whose ambition was to create a ‘universal science of organization’. His legacy is to have started systems thinking, thinking about a system as the emergent property of the interaction among all the components of the system and not as mere aggregates of its parts. System thinking in biology is not really new. Indeed, ever since the early microscopists it was always realized that the sum is more than the individual parts. In physiology, Claude Bernard (1813–1878) had already realized the common purpose of diverse physiological mechanisms to maintain homeostasis45. In molecular biology, however, such an integrative approach only became possible with the emergence of highthroughput technologies for measuring the large numbers of functionally diverse sets of elements in a cell with their patterns of selective interactions. Systems biology has now become an integrated approach to modeling biological systems in their entirety and to simulate their activity. For example, the human physiome project is to provide the framework for modeling the human body using computational methods46. The key challenge in these approaches is to distill the results of data-collection efforts into an interpretable computational form as the basis of a predictive model. A systems approach, then, involves repeated cycles of data collection and modeling. While systems biology is very far from computing the behavior of even a single cell, significant progress has been made. For example, models of heart function have already reached astonishing levels of detail, accuracy, and predictive power, as illustrated by realistic simulations of the beating of normal and abnormal hearts47. Studying the living world has brought us from the earliest microscopical observations of cells through the unraveling of the DNA sequence of hundreds of species to increasingly extensive collections of cellular constituents. The question then became how to convert this information into knowledge about the organism. Functional genomics and systems biology show great promise in becoming the centerpoints for exploring functionality in a quantitative manner, from the level of the genome and transcriptome to the physiology of organs and whole organisms. Can we use these same approaches to unlock the secrets of aging? And will that give us the means to develop interventions to ultimately halt its devastating effects?
INTRODUCTION
23
1.4 The causes of aging: a random affair What is aging? For practical reasons aging can be defined as a series of time-related processes occurring in the adult individual that ultimately bring life to a close. Aging is the most complex phenotype currently known and the only example of generalized biological dysfunction. Its effects become manifest in all organs and tissues. Aging influences an organism’s entire physiology, impacts function at all levels, and increases susceptibility to all major chronic diseases. Organ systems communicate with each other in order to maintain homeostasis. We need to decipher how these communications change over the life course and which cells and biological macromolecules are involved in such changes. It is therefore obvious that a systems approach is required to address the core problem of biological aging: the loss of homeostasis. Indeed, a comprehensive explanation of how we age requires an understanding at all levels of the decline of the many complex functionally interacting subsystems of an organism. Such insight should provide us with a rational basis for tracking aging processes from their downstream manifestations to the primary causes. Depending on what those causes are, this may in turn permit the identification of novel molecular and cellular targets for prevention and treatment of aging-related illnesses through pharmacological means. As we have seen, in the history of biology the discovery of the logic of life was followed by an understanding of the logic of aging. Following Weismann’s original non-adaptive concepts of explaining aging, most researchers now accept that aging is ultimately due to the greater relative weight placed by natural selection on early survival or reproduction than on maintaining vigor at later ages. This decline in the force of natural selection with age is largely due to the scarcity of older individuals in natural populations owing to mortality caused by extrinsic hazards (Chapter 2). By contrast, our understanding of the proximal causes of aging is limited. One can argue that this is due in large measure to our inability in the past to study aging systems. Instead, ample information has been gathered about individual cellular components at various ages, but this has not allowed a clear understanding of the integrated genomic circuits that control mechanisms of aging, survival, and responses to endogenous and environmental challenges. With the emergence of functional genomics and systems biology, we finally have the opportunity to study aging in a comprehensive manner, as a function of the dynamic network of genes that determines the physiology of an individual organism over time48. Increased technological prowess increases confidence levels. Our increased capacity to handle complex problems in biology whetted appetites for knowing the pathways that control the gradual changes in the structure and function of humans and animals that occur with the passage of time and their relationship with functional decline, disease, and death. In the past, aging was not always considered a serious biological problem. In contrast to a disease, aging was thought to be inevitable, with attempts to intervene in its many adverse effects better left to charlatans trying to interest the public in anti-aging
24
INTRODUCTION
products that often lack any rational basis. Perhaps in part because of unusual difficulties in studying aging, its science was seen by many as less rigorous and mainly phenomenological. Nevertheless, it was in the days when aging research had a low profile that some major scientific minds laid the groundwork for our current understanding of why and how we age. These individuals, true giants who laid the foundations for the science of aging, include George Sacher (1917–1981), Nathan Shock (1906–1989), Bernard Strehler (1925–2001), Alex Comfort (1920–2000), John Maynard Smith (1920–2004), Zhores Medvedev (London, UK), Paola Timiras (Berkeley, CA, USA), Leonard Hayflick (San Francisco, CA, USA), George Martin (Seattle, WA, USA), the previously mentioned Denham Harman, and several others49. Whereas charlatans and their anti-aging products have by no means disappeared, studying longevity and aging has now become respected and its accomplishments frequently evoke great enthusiasm, as exemplified by an increasing number of high-profile publications and abundant interest from the respected lay press. This development was greatly aided by the discovery, originally in the laboratory of Tom Johnson (Boulder, CO, USA) in the early 1980s, that a single gene mutation, called age-1, dramatically extended lifespan of the nematode worm C. elegans50. This was important because at that time it was generally believed that aging was too complex to be significantly delayed by altering a single gene. This lack of single genes affecting the normal aging process made the field unattractive for scientists who were used to relying on consistent effects of a limited number of genes involved in specific mechanisms, related to developmental or disease processes. With the discovery of the first gene affecting lifespan, aging had taken its place as a problem that could be addressed by studying the coordinated action of the products of multiple genes, similar to differentiation, development, and disease. In other words, aging had become a worthy object of study and began to attract highly reputed, well-funded scientists. At this point in time, there are hundreds of mutant genes in a variety of organisms, from yeast and fruit flies to mice, which increase longevity by dampening growth, reproduction, energy metabolism, or nutrient sensing. There are also mutant genes that cause accelerated aging, but these are not always generally accepted due to the difficulties in demonstrating that reduced longevity is genuinely due to accelerated aging or merely a result of a disease or developmental defect. This is discussed extensively in Chapter 5. Nomenclature for genes affecting aging is confusing. Analogous to disease genes— genes that cause disease when mutated—genes with mutations causing increased longevity are sometimes called longevity genes. Since others speak of gerontogenes and yet others of aging genes, this leads to difficulties in understanding the normal function of the pathways in which these genes act. Since the pathway in which age-1 and other longevity mutants in the nematode acts is really pro-aging, I will consistently call such genes aging genes. Longevity genes are genes encoding cellular processes that protect the organism against toxic insult. I realize that this departs from the disease gene
INTRODUCTION
25
nomenclature, but since this was wrong in the first place there is no need to also make the same confusing mistake for aging. As described in the next chapter, the novel approaches of functional genomics and systems biology greatly facilitate the further unraveling of the functional modules of longevity control. Along these lines rapid progress is now being made in understanding the pro-longevity responses that result from dampening the activities of aging genes. It is likely that very soon the mechanisms of action of such genes and the network of interactions that gives rise to increased longevity will be resolved. The problem of aging, however, has two faces and the specific programmatic responses extending lifespan is only one of them. Behind the other face of aging is stochasticity, an important aspect of biology in general, but often ignored since random variation is difficult to capture even with our current, highly sophisticated, biological tool set. While programmatic responses may defend us against toxic insults, it may be stochasticity that is behind the proximal cause of aging. For some time, evidence has been emerging that aging is caused by the accumulation of cellular damage, a random process. The programmatic component of aging may merely control the rate and extent of damage accumulation through dampening growth and reproductive processes with damage production as their inevitable side effects, or by upregulating processes of cellular defense. As we shall see in the next chapter, organisms are able to manipulate the allocation of their resources and balance reproductive efforts against somatic maintenance and repair. For some species, such as C. elegans, this flexibility is so high that interference in pathways of growth and reproduction, for example through single-gene mutations, can lead to 6-fold increases in lifespan. However, such dramatic effects are unlikely to occur in mammals due to their much greater complexity. Hence, whereas functional-genomics approaches can help us to more fully understand how lifespan is controlled in a variety of organisms, for studying aging it will be necessary to focus on its proximal cause: the accumulation of somatic damage. In this respect, there is now ample evidence that damage to the genome can explain many of the most important phenotypes of aging. This book is focused on the possibility that the genome is both the creative engine behind longevity, as this emerged during evolution in ever more robust manifestations, and the main target of the somatic damage that ultimately limits life. Since the original emergence of a genome 3–4 billion years ago, there has been a divergence into the current estimate of 30 million genomes, each representing a unique species. Evolution by natural selection requires the occurrence of mutations. If beneficial, such mutations are perpetuated. It is through mutations that Darwinian selection could lead to increasingly complex genomes and the adaptation of their hosts to the various challenges of a continuously changing environment. Because they occur at random, most mutations have adverse effects. During evolution, such genomes fall by the wayside as a consequence of natural selection.
26
INTRODUCTION
Perpetuation of life Evolutionary diversity
M
Individual development
M
M
M
Aging?
RNA
Protein
Function
Fig. 1.5 The genome harbors all the instructions for providing function to the somatic cells of an organism through RNA and protein. Random alterations in its information content in the germ line drive evolutionary change, whereas similar changes in the somatic cells could be the cause of aging. Mmutation.
Genomes are not only subject to variation in the germ cells transmitted by parents to offspring. New variation accumulates also in the soma through mutation. Such instability of the somatic genome is much more extensive than originally thought in the immediate aftermath of the double helix. As we shall see later, organisms gradually move from having cells with very similar genomes towards mosaics of cells each with their own genotype. It is the thesis of this book that the same time-dependent instability of genomes that gives rise to evolutionary diversity also leads to aging at the somatic level. In both cases it can do that by randomly influencing the mechanisms by which genomes provide function (Fig. 1.5). In the following chapters I will describe the logic of random genome damage as a major, highly conserved proximate cause of aging, discuss the ongoing efforts to unravel the many factors that determine this gradual loss of genome integrity of somatic cells, and critically review the evidence that genome alterations cause aging and its associated phenotypes, in the context of possible alternative explanations. In the final chapter I will discuss the impact of genome deterioration on our options to delay, halt, or even reverse the process of aging in mammals.
2
The logic of aging
In the eyes of many, the science of aging must look like a quest for the Holy Grail, a confusing series of contradictory approaches towards some elusive object. Today the field can roughly be divided into three main branches (Fig. 2.1): the biometric branch, seeing aging as infinitely complex and hardly amenable to intervention; the inductive branch, which attempts to explain aging in terms of few relatively simple, universal mechanisms; and the regeneration and renewal branch, with its focus on replacement and remodeling. The early idea that aging is caused by the accumulation of mutations in the genome of somatic cells is clearly part of the simplistic branch and stems from the time when it was popular to explain aging as theories of a single cause51. Although such unitary theories have recently undergone some revival with the discovery of conserved pathways regulating longevity in multiple species, from invertebrates to mammals (discussed further below)52, they lose a lot of appeal when confronted with the enormous complexity of the aging process as it takes place in different species53. Nevertheless, explaining aging in terms of one universal, driving force is not entirely unrealistic in view of the universality of the principles behind the living world on this planet. In spite of its bewildering variability, all life is based on combinations of the same, relatively few molecular species— sugars, fatty acids, amino acids, and nucleotides—and the single, universal, organizing principle of a genome that perpetuates itself as sequences of nucleic acid but functions by being expressed in the form of protein. Is it possible that aging has its own inherent logic driven by a teleological process towards a specific goal? This is not to say that as a biological process aging could have some cosmic purpose. Such matter is beyond the scope of this book and beyond the scope of the biological sciences. However, whereas questions of what and how are often satisfactory in physics, biology is full of goal-directed processes guided by programs. Good examples are differentiation, development, and certain behavioral patterns. The question, then, of why we age—that is, the evolutionary causation of aging as distinguished from its proximate causes—is a legitimate one and should be actively pursued. Indeed, in the past it is this type of question that has led to the most important discoveries in biology. Whereas aging is different from most biological processes, in the sense that it resembles mostly random degeneration, which is unlikely to be programmed, it could still have its own logic, hidden in the depth of the history of life. If aging is inevitable, what then are the characteristics of a genome to let that happen?
28
THE LOGIC OF AGING
Regeneration newal Bra d Re nc an h
n tio ta
ve
if y
Br
a
In
S
du
t
ic
cti
pl
Na
etr
e
Science of Aging
im
t in
in g
Los
Bio m
ur
nc
Branch
N ature
ing on Cl
Transpla n
planting Natur e Sup
h
Fig. 2.1 The science of aging as a search for the Holy Grail: three different, parallel approaches, often difficult to reconcile.
2.1 Aging genes One of the reasons why August Weismann’s original idea of aging as an adaptive trait, which evolved to get rid of old, worn-out individuals (Chapter 1), makes little sense is the unlikely event of meeting such an organism in the wild. Whereas in nature animals do undergo aging, at least the early stages of the process, few if any of them ever reach the advanced state of decrepitude experienced in protected environments, such as by laboratory animals or humans in advanced societies54. Animals in the wild are likely to die early, from predation, starvation, or accidents, before the ravaging symptoms of aging can become manifest55. There are, therefore, very few animals to experience the advanced stages of aging in their natural habitat. In the past, the same used to be true for humans, only a selected few of which had the good fortune to reach old age. This lack of expression of the aging phenotype in nature makes it an unlikely target of natural selection. Hence, there are no genes that cause aging like there are genes that specify development and maturation.
THE LOGIC OF AGING
29
That aging is nevertheless a real biological phenomenon common to most animal species can be derived from the observation that under protective conditions individuals of the same species, which in the wild never have the opportunity to get old, now reveal all the phenotypic characteristics of increased structural and functional degeneration, ultimately causing their death. The aging phenotype and the similarities of many of its characteristics within and among species will be discussed in some detail in Chapters 5 and 7. Here I will only point out the main reason why we believe that aging is a genetically controlled process of intrinsic degeneration and not just a series of accidental changes over time. If aging as a biological process did not exist, one would expect the probability of death to stay the same with advancing age. However, in protective environments with the threat of external mortality effectively minimal, the probability of death increases exponentially with age once maturity is reached. This was first discovered by the British actuarian Benjamin Gompertz (1779–1865) in 1825, when the average lifespan of humans had just begun its dramatic rise. Gompertz showed that after about age 35 the probability of death doubles every 7 years. Of note, while this increase in mortality has been observed in all species that undergo aging, at very old age it no longer seems to hold. This has been most convincingly demonstrated for invertebrate species, large numbers of which can be studied relatively easily. Large-scale studies of the lifespans of tens of thousands of medflies, fruit flies, or nematode worms have shown a deceleration of mortality at old age, which may in fact also be true for humans56. The slopes of the Gompertz plots (Fig. 2.2a) reflect the age-specific death rates (effectively the rate of aging) and can be converted into survival curves (Fig. 2.2b). When comparing protected populations with non-aging populations in the wild, the Gompertz plots and the survival curves reveal dramatic differences. The mortality curve for wild animals no longer shows the exponential increase with age (because death is accidental and independent of age) and the survival curve is not rectangular but concave in shape. It would apply to animals that are subject to high rates of external mortality from predation and other environmental factors. For most animals in the wild the situation is probably somewhere between these extremes. The survival curves reveal both the average lifespan of the population (the age at which 50% of the cohort has died) and the maximum lifespan, which is essentially the age of the last survivor or the average age of the longest-lived percentile or decile. The average lifespan or life expectancy is determined to a large extent by external factors. It can differ dramatically between the protected environment and the wild. The maximum lifespan is much less dependent on environmental conditions and is basically a function of the genetic make-up of the species. This explains why it does not change much when conditions improve. Careful comparisons of maximum lifespans for different animal species in captivity reveal significant differences, which indicate that lifespan is under genetic control (Fig. 2.3).
30
THE LOGIC OF AGING (a)
(b)
Wild (no aging)
100
Survival (%)
Log mortality
Wild (no aging)
Protected (aging)
Protected (aging)
Mean lifespan 50 Maximum lifespan
Age
Age
Fig. 2.2 (a) So-called Gompertz plot for animals in the wild (dashed line) and for animals under protective conditions (solid line). Both plots start in early adulthood. (b) The survival curves corresponding to mortality patterns in the wild (dashed line) and under protective conditions (solid line). Aging is revealed by the rectangularization of the survival curve. 140 Human 120 years
120
Lifespan (years)
100
80 Chimpanzee 55 years
60
Rhesus monkey 35 years
40
20
Squirrel monkey 20 years Mouse 3 years
0 Species
Fig. 2.3 Maximum lifespans for different animal species under protective conditions.
It should be kept in mind that lifespan values, especially maximum lifespans, are difficult to determine for animals in the wild because of difficulties in monitoring birth cohorts and the limited opportunities to survive beyond the age of reproduction. Mice, for example, may only live for a few months in the wild; they generally die early from predation
THE LOGIC OF AGING
31
or cold. For humans survival curves can be generated from census mortality tables. However, such data are only available for the last couple of hundred years and even then not always fully reliable. Hence, although it is often argued that maximum lifespan is fixed this is really not supported by strong data. In fact, there is evidence, for example from the observed deceleration of mortality at extreme old age mentioned above, that lifespan, even maximum lifespan, is more fluid that we tended to think (see Chapter 8 for a more detailed discussion of these issues). At a population level, then, aging can be defined as a process that begins after maturity and results in an increasing probability of death. It is accompanied, at the individual level, by a gradual impairment of bodily functions, a host of structural changes and an increased incidence of certain diseases. In principle aging can include positive alterations and some have even defined it as all possible changes in an organism between conception and death. To indicate aging-related adverse changes exclusively some authors use the word senescence, which can be defined as those irreversible, deteriorative changes causing functional decline, disease, and death in aged organisms. This is somewhat confusing since, as we will see, senescence is also defined as the irreversible loss of proliferative activity of a population of cells, for example, yeast, protozoa, or mammalian cells. It is also used to simply indicate the aging of cells. In this book I will frequently use the word senescence in all these meanings. For more extensive discussions of aging in the animal world I refer to some outstanding works by some of my colleagues who are much better versed in the basic aspects of aging as a biological phenomenon. Examples are the monumental Longevity, Senescence and the Genome by Caleb Finch57, Why We Age: What Science is Discovering about the Body’s Journey Through Life’ by Steve Austad58, and Time of Our Lives: the Science of Human Ageing’ by Tom Kirkwood59. The question has arisen as to how aging, which in evolutionary terms can be simply defined as the decline in fitness after the period of first reproduction, could ever emerge as a distinct, albeit complicated, phenotype. Indeed, direct natural selection should favor the suppression of senescence rather than its promotion. That is, the accidental inactivation of a gene causing a program of aging would provide an immediate selective advantage to its carrier, resulting in a rapid spread of immortality. In the next sections I will briefly explain why this is not the case and how aging may have evolved as a natural consequence of the logic of life.
2.1.1 AGING IS A BY-PRODUCT, NOT A GOAL OF NATURAL SELECTION It is now generally accepted that the age-related decrease in fitness is non-adaptive; that is, it is not controlled by a purposeful genetic program similar to development. As we have seen in Chapter 1, August Weismann was the first to propose a non-adaptive
32
THE LOGIC OF AGING
explanation for aging. While, as just discussed, his name is generally associated with the idea of aging as a beneficial trait—evolved during evolution to cleanse the population of old, worn-out individuals, consuming resources without being reproductively active— Kirkwood and Cremer7 pointed out that a little later Weismann also developed a nonadaptive explanation of aging. He recognized that if a specific function loses its usefulness, for example, due to a change in the organism’s environment, natural selection will start ignoring it and the character will degenerate and disappear. Being ignored by natural selection is probably exactly what applies best to the supposedly useless period of life when reproductive duty is fulfilled and the theoretical chances of surviving much longer have become slim. Aging would therefore be the logical result of the declining force of natural selection after the period of first reproduction. Weismann, therefore, understood that aging by itself is unlikely to have an advantage, which contradicts his earlier idea of a beneficial cleansing mechanism. The non-adaptive concept of how aging could originate in nature was first discussed systematically by Peter Medawar in the 1950s60, going back to ideas of Fisher, Haldane, and Hamilton10. Medawar proposed that aging was the necessary result of constitutional mutations, accumulated in the germ line over evolutionary time, that reduce fitness late in life. As mentioned above, in the wild only a small fraction of a birth cohort will reach advanced age while continuing to reproduce. Therefore, the later the adverse effects of such mutations manifest after the period of first reproduction, the less likely that they will be weeded out by natural selection. The frequencies of such alleles, especially in small populations, will then drift randomly from generation to generation. In this way, natural selection acts to postpone adverse actions of genes until late age, when under normal conditions no animals can be expected to be alive anymore. Under such conditions natural selection does not ‘see’ the effect of these mutations, which only show up under more optimal conditions, such as in captivity. In general, natural populations experiencing low mortality, from predation or disease, will postpone late adverse effects further than populations of the same species in a habitat with high extrinsic mortality. Indeed, under conditions of high extrinsic mortality, lifespan would already be short, pushing back the reproductive period and removing any evolutionary incentive to cleanse the genome from random mutations with adverse effects after this period. In captivity, under optimal conditions, such animals would display shorter lifespans. In other words, their intrinsic mortality would adapt to their extrinsic mortality. Adaptation of intrinsic mortality to extrinsic mortality can also work the other way around; that is, lifespan of a population of organisms can be extended by artificially extending the period of reproduction. This has been demonstrated, for example, in outbred populations of D. melanogaster, the fruit fly. Clare and Luckinbill61 restricted reproduction of this organism to late age, thereby increasing the intensity of selection during the later portion of the lifespan. They did this for 21–29 generations, at two different
THE LOGIC OF AGING
33
larval densities. Populations with high and uncontrolled numbers of competing larvae responded strongly to selection for late-life reproduction, with the length of adult life increasing by as much as 50%. Under such conditions selection produced true-breeding long-lived lines. When populations of developing larvae were held low, however, longevity fluctuated wildly during selection, showing little overall response. Interestingly, this experimental design allowed the simultaneous demonstration of the existence of natural pro-longevity gene variants and the strong influence of the environment on the selection of these variants. Later, the prediction that higher extrinsic mortality rates lead to the evolution of shorter lifespans in Drosophila was confirmed by directly comparing populations of wild flies under conditions of high and low imposed mortality62. Evidence that different levels of environmental hazards dictate longevity through evolutionary selection has also come from studies of vertebrate animals in the wild. Observations by Steve Austad (San Antonio, TX, USA) on two opossum populations, one on an island not subject to predation and one on the mainland where they were exposed to significant predation, confirmed the prediction that the island population had the highest longevity63. Environment is also not the only factor that appears to play a role. In general, attributes that reduce extrinsic mortality, such as wings (birds, bats), protective shells (turtle), or brains (humans) are generally associated with increased longevity57,58. However, the situation is not straightforward and other factors, such as ecology, play important roles. The aforementioned selection of long-lived flies through late-life reproduction only worked at high larval density. In a study of guppies derived from environments with high and low mortality rates, lifespan was found to be longest in streams where predators co-occur with guppies and shorter in individuals reared in the upper reaches of streams by waterfalls, from which predators are often excluded64. There are several possible explanations for this unexpected finding, including a reduction in population size of the guppies by the predators, which would result in an increase of the abundance of food and other resources leading to increased survival65. It should be noted that this (thus far) isolated finding does not falsify the evolutionary theory of aging. It merely illustrates that unlike the situation in the physical sciences theories in biology are often ambiguous and have no impenetrable mathematical basis. This is necessarily so when starting from the shared organizational features manifested by organisms. It is not possible in biology to reject theories on the basis of Popper’s falsification principle. In biology exceptions are often the norm. This does not make biology unphysical. It simply means that in biology generalizations are often based on inadequate or incomplete theories. The correlation between high extrinsic mortality and short lifespans may be a consequence of the trade-off most organisms face between investment in somatic maintenance and reproductive effort. This was first recognized by Tom Kirkwood and formulated in his disposable soma theory of aging9, already mentioned in Chapter 1. This theory predicts that high extrinsic mortality would favor investment of scarce resources in early
34
THE LOGIC OF AGING
reproduction rather than somatic maintenance and repair, which would not be required for a population with a low risk of survival in the wild. This of course implies that agerelated somatic degeneration and death is caused by the accumulation of unrepaired somatic damage, a reasonable explanation, which is now supported by a large body of evidence. For example, the extended longevity in late-reproducing fruit flies mentioned above appeared to be associated with a coordinated upregulation of genes specifying cellular defense mechanisms against free-radical attack. Consequently, these flies showed decreased levels of oxidative damage to proteins and lipids66. In addition, delayed senescence in the long-lived flies was accompanied by reduced fecundity, supporting the idea of a trade-off between reproductive effort and survival (see also below).
2.1.2 AGING OF UNICELLULAR ORGANISMS From an evolutionary point of view, therefore, age-related decline and death is the effect of genetic variants (aging genes) that have escaped the force of natural selection by acting only post-reproductively. Ultimately, this is due to the separation of soma from germ, which creates a life cycle in which the soma is reconstructed in every generation from a germinal blueprint. As we have seen, reconstruction of the soma (or reproduction) is greatly favored by natural selection, especially under conditions of high external mortality. The fate of the old soma is not a priority. In essence, this means that the deleterious effects of mutations accumulated in the germ line are shed to the parents. However, this implies that unicellular organisms should not age because they do not have a distinction between the germ line, which is to be preserved, and the soma, which is expendable. Although for most unicellular species (as well as some metazoa without a clear distinction between germ line and somatic tissue, such as Hydra) there is indeed no evidence of aging57, certain unicellular eukaryotes, such as budding yeast and protozoa, undergo a process of cellular senescence after a given number of population doublings. In such cases there usually is asymmetric division and it has been suggested that in parallel to the soma of multicellular organisms, organisms such as yeast segregate their senescent changes to the mother cell. Hence, like in most metazoa, the negative effects of aging are confined to the parent. A similar situation has been observed in asymmetrically dividing prokaryotes67,68. Even in a symmetrically dividing organism, like E. coli, which divides by binary fission into seemingly identical siblings, aging has recently been demonstrated. This was attributed to inheritance of the old pole; that is, the end of the cell that has not been newly created during division, but is pre-existing from a previous division. Old poles can exist for many divisions, and by following repeated cycles of reproduction by individual cells, through automated time-lapse microscopy, it was shown that the cell that inherits the old pole exhibits a diminished growth rate, decreased offspring production, and an increased incidence of death68.
THE LOGIC OF AGING
35
Protozoa have been studied extensively in the context of aging and reproduction. In his beautiful book, Sex and Death in Protozoa69, Graham Bell (Montreal, Canada) has outlined in intricate detail many if not all key concepts of aging in unicellular organisms in comparison with multicellular ones. Protozoa reproduce asexually, by fission, and sometimes also sexually, which is termed conjugation. Although protozoa do not divide asymmetrically, they have a life cycle: immaturity, maturity, and senescence69. In a protozoan clone, senescence occurs in the absence of conjugation. Conjugation involves fusion of two paired individuals, meiosis of the diploid nucleus (termed the micronucleus as distinct from the macronucleus, which disintegrates) and exchange of one gamete nucleus. After fusion of gametes the diploid state is reconstituted and new individuals are generated. Conjugation prevents senescence, which only occurs in its absence. The zoologist Tracy Morton Sonneborn (1905–1981) discovered in the early 1950s that autogamy, the union of gametic pronuclei within the same cell, without conjugation, could substitute for conjugation in preventing senescence in protozoa70. Bell has interpreted senescence in protozoa as being caused by the irreversible accumulation of deleterious mutations in isolate lines and not as some adaptive response. However, it is possible that senescence offers a selective advantage to populations of protozoa. Indeed, Cui et al. used computer simulation to analyze two hypothetical species of protozoa, one with and one without senescence71. The two species were subject to the same rate of deleterious mutation and could undergo fission or conjugation. Conjugation was assumed to re-set the clock, whereas senescence would eliminate an individual reaching its age limit. The results of the simulation indicated that far fewer recessive deleterious mutations had accumulated in the species employing senescence. The authors speculated that senescence had emerged as an effective way of interrupting the pathway of mutation accumulation increasing the chance of sexual reproduction to produce high-fitness offspring. In this case, senescence has been explained as offering an advantage in combination with sexual reproduction. This would be in keeping with the more general idea that aging is the price we have to pay for sex, which is almost certainly false because, as outlined above, aging is not a major cause of mortality in natural animal populations72. However, senescence in protozoa is not necessarily identical to aging in mammals and it is certainly possible that the process has some adaptive value (see further below). It should be noted that Bell argued against senescence in protozoa as an adaptive response since the onset of senescent changes is highly variable between independent cultures in contrast to the onset of sexual maturation, which is invariant. The basis of any discussion of mutation accumulation is the realization that in living organisms genetic information cannot be transmitted without error. In a finite asexual population deleterious mutations will accumulate over time because sooner or later those cells harboring the least mutations will be extinguished and cannot be restored. This phenomenon, called Muller’s ratchet73, is responsible for senescence in isolated cultures of protozoa and sets a limit to the longevity of germ lines. Sex would then act as a repair
36
THE LOGIC OF AGING
device for acquired mutations by re-shuffling genomes, effectively resulting in some genomes with low mutation loads. As explained by Bell, while this protection would be most effective during cross-fertilization, also autogamy will offer some protection, which explains the result obtained by Sonneborn, mentioned above. It should be realized that population extinction by Muller’s ratchet (germinal senescence) is essentially different from the concept of aging genes resulting from the accumulation of mutations in the germ line. As pointed out by Bell, the germ line ages because it accumulates mutations adversely affecting fertility. This can effectively terminate a population. The soma of multicellular organisms ages because selection will not cleanse their genome of genes with adverse effects on the soma at late age. Nevertheless, the ratchet would operate in all genomes over time and there is no reason to assume that it would not operate in somatic cells of metazoa, especially stem cells which would readily develop into clonal lines creating mosaics for the mutations which have arisen. This situation would be analogous to senescence of a protozoan clone, but significantly more complicated in view of the interdependence of different types of somatic cells in metazoan species. In summary, Medawar’s idea of aging as a result of the accumulation in the germ line of mutations creating gene variants with adverse effects on the soma late in life, after the age of first reproduction, is a logical explanation for the sheer universality of aging in the animal world. Such a random process of evolving aging genes suggests that aging is affected by many genes, each with a relatively small effect. This would make aging infinitely complex and different from species to species. While aging is indeed characterized by randomness, with variation within and between species the rule, this explanation is still unsatisfactory since it does not address the great similarities in the types of aging processes that are also a characteristic of aging in animals.
2.2 Pleiotropy in aging The evolutionary logic of the Medawar theory begs the question of the type of aging genes we should be looking for when trying to identify the mechanisms that control aging and longevity. Before discussing this it is important to clarify what we mean by aging genes. First, Medawar’s accidental mutations in the germ line are not necessarily in genes, but could be in DNA sequences that control the expression of genes. Second, as should be clear from the discussion above, there are no genes specifically selected to cause aging. Instead, most aging genes are variants of genes or DNA sequences that have adverse effects after the age of first reproduction. Finally, as discussed in Chapter 1, genes do not act alone, but in functional modules, which arise from interactions among components in the cell (proteins, DNA, RNA, and small molecules). This context should be kept in mind every time the term aging gene is used.
THE LOGIC OF AGING
37
George Martin has distinguished private from public aging genes74. Private aging genes find their origin in Medawar’s accidental germ-line mutations discussed above, which are neutral at early age, but start exerting their adverse effects later, after the period of first reproduction. In the absence of selection, the frequency of such mutations in the population would be determined by genetic drift. They will, therefore, be relatively rare and can also be expected to be specific for the population or species in which they arise. Hence, they were termed private aging genes. J.B.S. Haldane predicted such genes in the 1940s based on his observation that Huntington’s disease, a genetic disorder, occurs late in life, after most people have already had their children. Apart from Huntington’s disease, in humans private aging genes could explain familial forms of age-related disease phenotypes, such as Alzheimer’s disease. Martin systematically analyzed known loci involved in human genetic disease, which had been cataloged by Victor McKusick (Baltimore, MD, USA) as Mendelian Inheritance of Man (a phenotypic companion to the HGP and available online as OMIM; www.ncbi.nlm.nih.gov/ entrez/query.fcgi?dbOMIM). Based on these phenotypic descriptions, Martin concluded that private mutations at a very large number of loci could play a causal role in the senescent phenotypes as they are observed among elderly75. In contrast to private aging genes, public aging genes are hypothesized to have arisen under the influence of natural selection, not because they cause aging but as a consequence of an early, beneficial effect. Their adverse by-products would only become manifest at later age. This involves the concept of antagonistic pleiotropy and was originally proposed by George Williams76. Since we know that genes usually have more than one effect (hence the term pleiotropy) it is more than likely that the same applies to the gene variants arising as accidental mutations. Since most mutations are bad, it would be unlikely that in the rare case of a beneficial effect of an accidental germ-line mutation early in life there would not be an adverse effect later. In this view, aging would be the result of the harmful side effects of genes selected for advantages they offered during youth. It is possible that there are many such trade-offs between early and late effects and, in contrast to Medawar’s deleterious mutations, which are neutral at early age, pleiotropic mutations could involve conserved mechanisms of universal benefit. Indeed, mutational variants of genes affecting similar functions early in life could have arisen independently in different populations, and even in different, only distantly related species. For this reason one can refer to such evolutionarily conserved genes as public aging genes. A few theoretical examples underscore the existence of the trade-offs dictated by public aging genes. To illustrate his pleiotropic gene theory of aging, Williams himself provided the first example, in the form of an arising gene variant with a favorable effect on bone calcification in the developmental period with the adverse side effect of depositions of calcium in the arterial walls at later ages76. As noted above, as an unfavorable character, senescence would always evoke direct action of selection to oppose it. Hence, additional gene variants would likely arise to suppress the adverse phenotype of calcium deposition in arterial
38
THE LOGIC OF AGING
walls (if environmental constraints permit). However, this mechanism would never succeed in suppressing the adverse trait completely due to diminishing selective pressure with age. Another example of a trade-off that may explain human aging-related disease is inflammation. Genetic pathways specifying a powerful response to infection have undoubtedly a high survival value, but may at later age contribute to disease-causing degeneration, such as atherosclerosis77. One could think of a large number of trade-offs in gene action, reflecting on agerelated disease phenotypes. Cancer is one of them. Tumor-suppressor genes are obviously advantageous for young organisms and mutations inactivating such genes would not be passed on to many offspring. Apoptosis, or programmed cell death, and cellular senescence, irreversible mitotic arrest, are both critical processes for suppressing tumorigenesis in mammals78. Both responses are highly conserved in eukaryotic organisms also for reasons other than suppressing cancer. Apoptosis, for example, plays an essential role in embryonic development and also later in maintaining normal tissue homeostasis. As first realized by Judith Campisi (Berkeley, CA, USA), in mammals both apoptosis and cellular senescence are likely to be antagonistically pleiotropic, since they help to suppress cancer at early age, but possibly at the cost of promoting aging at later ages by exhausting progenitor- or stem-cell reservoirs79. This possible antagonism between cancer and non-cancer degenerative aging phenotypes will be discussed in more detail in Chapter 7. Another example of antagonistic pleiotropy could be the production of ATP by oxidative phosphorylation, the main pathway of energy production. This effective method of harnessing energy arose early in evolution and has been conserved with relatively minor variation. However, cells produce reactive oxygen species (ROS) as by-products of oxidative phosphorylation and I already mentioned in Chapter 1 that it was Denham Harman who originally hypothesized that such free radicals are one of the major factors responsible for the aging of cells23. Since the accumulated damage from ROS usually only becomes manifest at late age, this could be another example of postponing adverse effects of a beneficial pathway. Like other processes of a conserved beneficial nature, genes controlling oxidative phosphorylation can be considered as public aging genes. Probably the most ancient, hypothetical example of antagonistic pleiotropy is the complex of intertwined pathways to replicate, recombine, and repair DNA or, originally, RNA. While these systems guarantee life’s continuity, there is a price in the form of the doubleedged sword of mutations, the result of erroneous DNA transactions. Mutations are necessary as the ultimate source of genetic variation upon which evolution depends, yet most of them are harmful. We have already seen that mutations with late-life adverse effects but which are neutral at early age are likely to accumulate in the germ line. A high rate of germ-line mutation can lead to population extinction, which may have been the driving force behind the evolution of sex, mate choice, and diploid life cycles, all ways to limit the adverse effects of mutations. Since the adverse effects of mutations in somatic cells would only become manifest at late age, selection to improve the accuracy of DNA metabolism far beyond the age of first reproduction is absent.
THE LOGIC OF AGING
39
Antagonistic pleiotropy can also act directly on reproductive success (sexual selection) rather than on the ability to deal with environmental factors (survival selection). Indeed, lower fecundity is often found associated with longevity, such as in the aforementioned insular opossum population experiencing the least extrinsic mortality63. The importance of sexual selection is further underscored by a series of experiments reported by Sgro and Partridge80, the results of which elegantly demonstrated that antagonistic pleiotropy is likely to be more frequently responsible for aging phenotypes than Medawar’s neutral mutations. In long- and short-lived lines of fruit flies, generated by making these flies reproduce late or very early (see above), Sgro and Partridge initially observed similar mortality. However, after 30 days, the short-lived lines, with eggs laid by very young adults, experienced a wave of higher mortality, which did not occur in the flies selected to live for longer80. Interestingly, this wave of higher mortality, causing the early reproducers to age quickly, could be abrogated by the ablation of egg laying, for example through irradiation. Hence, rather than late, adverse effects of mutations neutral at early age, these results suggest a damaging effect of early reproduction. Selection for life extension could be the result of a switch in resource allocation, from reproduction to somatic maintenance. Alternatively, early reproduction could have caused the damage directly with the effects accumulating over time, to result in the wave of mortality. In spite of this elegant experimental example of antagonistic pleiotropy in action, with sexual selection as the possible driver, an inverse relationship between fecundity and longevity is complex and not always found81. In summary, aging genes are not considered to have emerged on the basis of some selected value for fitness but are merely by-products of evolution. They are thought to dictate trade-offs between beneficial effects early in life and adverse effects later. In the next section we will discuss several examples of such genes. An adverse effect of any aging gene can be suppressed by another gene, which we can call longevity genes or longevityassurance genes. It is conceivable that the emergence of aging genes and longevity genes go hand in hand, under the influence of environmental constraints. Although the existence of a large number of private genes, specifying aging mechanisms peculiar to certain species and to certain individuals within a species, cannot be ruled out, evidence points towards a major role of public mechanisms specifying highly conserved beneficial pathways, such as oxidative phosphorylation and apoptosis. This is in keeping with the many similarities of aging within and across species and, based on the ancestry of some universal pathways of life, opens up the possibility that aging has been with us from early times.
2.3 Interrupting the pathways of aging One way to identify aging genes and their mechanisms of action would be to screen a population for individuals harboring inactivating or weakening mutations in a gene controlling a pathway of aging. This would work well for Medawar’s gene variants that are
40
THE LOGIC OF AGING
neutral at young age and also for Williams’ pleiotropic gene variants when depriving the individual of an early beneficial effect that is not lethal under the conditions the animal is studied. The latter may be difficult in mammals with their complicated genomes harboring numerous gene–gene interactions that can easily lead to adverse side effects. As we will see, most of the aging genes thus far identified have indeed been found in relatively simple invertebrates, although large-scale genetic screens are of course also much easier to perform in short-lived simple organisms than in mice or rats. Whereas it was previously recognized that many single-gene mutants could shorten lifespan, a major increase in lifespan due to the effect of one gene was considered unlikely in view of the presumed multicausal nature of the aging process. This does not mean that mechanisms to delay aging and extend lifespan were not known. As described further below, the phenomenon of extending lifespan through calorie restriction has been known since the 1930s and the effect of temperature and reproductive status on lifespan of Drosophila was described as early as the 1920s by Raymond Pearl (1879–1940). Indeed, John Maynard Smith (1920–2004) was probably the first to describe an increase in lifespan due to the effect of a single-gene mutation in Drosophila. He found that the ovariless mutant (a mutant where the females entirely lack ovaries) as well as virgin or partially heat-sterilized females have extended survival compared to controls82. However, these results were not generalized in terms of the existence of many genes that could be manipulated to alter functional pathways controlling longevity. It is now well understood that single-gene mutants can affect a host of downstream processes and therefore significantly influence major prolongevity systems. This is the basis of the wealth of longlived mutants that have now been discovered in nematodes, fruit flies, and, more recently, in mice83.
2.3.1 THE NEMATODE WORM, C. ELEGANS The mutant gene that transformed this field by its demonstrated pro-longevity effect in the nematode was first identified in the laboratory of Tom Johnson (Boulder, CO, USA)50, who called it age-1. Later, this gene turned out to encode the C. elegans ortholog of the phosphoinositide 3-kinase p110 catalytic subunit, a central component of the C. elegans insulin-like signaling pathway, lying downstream of the DAF-2/insulin receptor (see below) and upstream of both the phosphoinositide-dependent protein kinase 1 (PDK-1) and thymoma viral proto-oncogenes 1 and 2 (AKT-1/AKT-2 kinases) and the DAF-16 forkhead-type transcription factor, whose negative regulation is the key output of the insulin signaling pathway84 (see below). Cynthia Kenyon (San Francisco, CA, USA) and co-workers subsequently demonstrated that mutations in daf-2 also conferred longevity to the worm85. This gene was later identified in the laboratory of Gary Ruvkin (Boston, MA, USA) as a insulin/insulin-like growth factor 1 (IGF-1) receptor homolog, thus acting (upstream) in the same pathway
THE LOGIC OF AGING
41
as age-186. Loss-of-function mutations in these genes and others acting in these pathways cause dauer formation, a state of diapause in response to food limitation and crowding. Weak mutations in such genes, however, allow these animals to become adults and live up to twice their normal lifespan. It appeared that the diapause-related genes confer their longevity-promoting effects through the action of a forkhead/winged helix transcription factor called DAF-16, identified independently by the Kenyon and Ruvkin labs87,88. DAF16 relocates from the cytoplasm into the nuclei of different cell types in the nematode and affects the activities of genes involved in many processes, including metabolism, stress response, and antimicrobial action89. Indeed, it is possible that the mechanism underlying the increased longevity of these and other mutants in different organisms involves the upregulation of longevity-assurance mechanisms (see below). The Kenyon lab also found that germ-line ablation increases the lifespan of nematodes90, similar to what has been found in Drosophila (see above). While this effect is independent of the upstream daf-2 gene, its signal does activate daf-16. DAF-16, therefore may be a master regulator integrating different longevity signals. Among genes conferring longevity on the nematode upon mutational inactivation or weakening, those that impact on mitochondrial function are especially frequent. Examples are the clk genes, so-called because they regulate physiological, developmental, and behavioral clocks during the nematode life cycle. Four clk mutants have been identified that show a moderate increase in lifespan91. CLK-1 has been shown to localize in the mitochondria of all somatic cells of the worm and is required for the biosynthesis of coenzyme Q992. Coenzyme Q plays a crucial role in the mitochondrial electron-transport chain, and clk-1 mutants rely on the Q8 homolog of Q9 synthesized by the worm’s diet of E. coli. Withdrawal of Q8 from the diet of wild-type nematodes extends adult lifespan by approximately 60%, probably by decreasing the release of free radicals from mitochondria or by increasing resistance to damage accumulation93. The first lifespan-extending mutations were discovered in nematodes in classical genetic screens, by mutagenizing worms to randomly disrupt gene function and subsequently recover long-lived mutants. However, with the emergence of the novel, highthroughput methods in functional genomics a more systematic approach became possible, which was quickly utilized by the main players in this by now highly competitive field. To identify in a systematic manner the different classes of genes that control C. elegans lifespan, both the Kenyon and Ruvkin labs began to apply the new RNAi approach (Chapter 1). Using so-called feeding RNAi libraries—that is, by feeding worms with bacteria expressing double-stranded RNA—over 17 000 nematode genes were screened by these investigators94,95. A number of known and unknown genes—as many as 89 in one of these studies—were found that extended nematode lifespan upon RNAi inactivation. Many of these genes turned out to encode components of the mitochondrial respiratory chain or were impacting mitochondrial function. Thus, the results obtained with this comprehensive RNAi screen confirm the idea that mutations that extend lifespan often involve genes participating in energy metabolism.
42
THE LOGIC OF AGING
2.3.2 THE FRUIT FLY, D. MELANOGASTER As expected from public aging pathways, when the interruption of a pathway leads to extended lifespan in one species, interruption of orthologous pathways in other species should give similar results. This has proved to be the case for the insulin/IGF-1 signaling pathway. As we have seen, reducing the activity of this pathway increases lifespan in the nematode. In Drosophila, complete inactivation of the gene that encodes the insulin receptor substrate (IRS) homolog (chico) extends lifespan by about 45%, but only in females. Heterozygous individuals of both sexes lived longer. There is also evidence that in chico mutants the longevity phenotype is associated with increased stress resistance. Similarly, hypomorph mutations in the insulin-like receptor (InR) gene extend the lifespan of females, but not of males52. In this case there is evidence that the longevity phenotype in the females is associated with infertility. Another aging pathway identified in Drosophila on the basis of mutants of increased lifespan is the target of rapamycin (TOR) pathway, which has now emerged as a major regulator of growth and size, as well as longevity96. Inhibition of this nutrient-sensing pathway, which can modulate insulin signaling and growth, extends lifespan in the fly in a manner that may overlap with caloric restriction; that is, a reduction in nutrient intake increasing lifespan across different taxa. Two partial loss-of-function mutations have been described in the fly, namely Methuselah (mth) and I’m not dead yet (indy), which increase lifespan by 30 and 100%, respectively. The effect is seen in both male and female flies without loss of fertility. However, the indy mutation only increases lifespan in the heterozygote flies. The mth gene encodes a G-protein-coupled transmembrane receptor97. In addition, mutations in the gene for its ligand, Stunted (sun), which encodes a subunit of ATP synthase, increase lifespan98. Whereas Stunted is found on the cell surface, the Methuselah–Stunted pathway may exert its role as a regulator of aging through the mitochondria. The indy gene encodes a dicarboxylate transporter, a membrane protein that transports Krebs-cycle intermediates in tissues participating in the uptake, utilization, and storage of nutrients. The life-extending effect of the Indy mutation may be due to an alteration in energy balance caused by a decrease in Indy transport function99.
2.3.3 CALORIE RESTRICTION Well before it was realized that single gene mutations could slow the rate of aging and increase lifespan in invertebrates, Clive McCay (1898–1967) reported in 1935 his unexpected discovery that rats that ate less lived nearly 30% longer100. Calorie or caloric restriction, as this is generally called, is underfeeding without malnutrition. Work from others, most notably a series of careful studies by Edward Masoro (San Antonio, TX, USA),
THE LOGIC OF AGING
43
subsequently demonstrated convincingly that reduced caloric intake rather than reduction of a specific nutrient was responsible for the beneficial effects, which also included a reduction or retardation of tumor formation and other aging-related phenotypes101. Importantly, a significant effect was still seen when calorie restriction in the mouse was initiated as late as 12 months of age102. Calorie restriction is unrelated to development since there was no effect when applied early, between 6 and 24 weeks in the rat103. Rodents on a calorie-restriction regimen have lower fasting levels of plasma glucose, insulin, and IGF-1. This suggests that decreased insulin signaling, similar to the situation in some of the longlived nematode and fly mutants, could also contribute to the calorie-restriction longevity effect. Life extension through caloric restriction has been observed not only in rodents, but also in various species of worms, flies, and yeast. Studies to test whether similar effects can be found in primates are still under way104. Calorie restriction is often assumed to be caused by a lower metabolic rate (rate of energy utilization), often thought to promote longevity. The idea that aging is inversely related to metabolic rate goes back to a publication by Max Rubner (1854–1932) in 1908 showing for five mammalian species that in spite of large differences in lifespan total metabolic output (the amount of energy consumed during adult life) per unit body weight was similar105. The conclusion was that animals with a high metabolic rate would have a shorter lifespan than animals with lower rates. Raymond Pearl (1879–1940) subsequently presented his rate of living theory based on survival experiments with Drosophila and cantaloupe seedlings106. His conclusions were that an organism’s duration of life is determined by its ‘inherent vitality’. Inherent vitality was defined by Pearl as the total capacity of the organism to perform vital action in the complete absence of exogenous energy; under conditions of complete starvation. Inherent vitality is then lost at a rate that equals the rate of energy expenditure. Lifespan, therefore, is inversely related to the rate of living. Naturally, it is tempting to link metabolic rate to natural free-radical production and the accumulation of somatic damage as originally conceived by Harman23. Indeed, there is ample evidence that calorie restriction retards the accumulation of oxidatively damaged molecules in aging rodents107. Whereas metabolic alterations are likely to play a role in both caloric restriction and the longevity phenotype of most if not all the long-lived mutants in worms and flies, variation in lifespan appeared not to be attributable to reduced metabolic rate. For example, whereas in rodents immediately after the initiation of calorie restriction metabolic rate rapidly declines, it is subsequently restored and stays the same per lean body mass, as in control animals108. In flies and worms, calorie-restriction-induced life extension is not associated with a lower metabolic rate and the same is true for the longevity phenotypes of the insulin-signaling mutants of these invertebrate species109–111. This does not rule out the possibility of metabolic alterations facilitating a reduced formation of reactive oxygen molecules.
44
THE LOGIC OF AGING
2.3.4 GENETIC MANIPULATION OF LIFESPAN IN MICE Naturally, genetic screens for longevity mutants are difficult to carry out in mammals because of their long generation times and expensive husbandry. On the other hand, they offer the enormous advantages that they are evolutionarily very close to humans and that lifetime phenotypic information on mice and rats is extensive (albeit not as complete as for humans). Studies of some spontaneous and engineered mutants suggest that mechanisms similar to those acting in nematodes and fruit flies also control longevity in mammals. Mice homozygous for mutations in the Prop-1 gene, termed Ames dwarf mice, are dwarfs and live 50–70% longer than wild-type mice112. These animals are deficient in growth hormone (GH), thyroid-stimulating hormone, and prolactin, and they show reduced levels of plasma insulin, IGF-1, and glucose. They also show a delayed occurrence of neoplastic lesions compared with their normal siblings113. Expression of the Pit-1 gene, which controls development of the pituitary gland, depends on expression of the Prop-1 gene. Pit-1-defective mice, termed Snell dwarf mice, are dwarfs and infertile like Ames mice, and they show a 40% increase in mean and maximum lifespan114. They also show a delay in collagen cross-linking with age and several age-sensitive indices of immune status. In Snell mice, evidence has been provided for reduced insulin/IGF-1 signaling in response to the GH deficiency115, which could be a key factor in the lifespan extension of these mice, as it apparently is in longevity mutants of nematodes and flies and, possibly, in calorie restriction. Of note, both natural and engineered GH-defective mice show increased longevity and have also very low circulating IGF-1114,116. Interestingly, like the longevity mutants in nematodes and flies, both dwarf mice and calorie-restricted mice display increased resistance to stress, pointing towards an upregulation of longevity systems (see further below). Reduced insulin signaling as a universal mechanism for delayed aging of mice is in keeping with the increased lifespan of female mice haploinsufficient for the IGF-1 receptor117, although this finding has thus far not been reproduced independently. Interestingly, both male and female mice with a disruption of the insulin receptor gene only in adipose tissue show a decreased body fat mass and increased longevity118. These mice live longer with normal caloric intake and retain leanness and glucose tolerance with age. While it is tempting to suggest that based on this finding it is leanness rather than metabolic changes that causes the extension of life, it should be realized that as argued by Masoro119 the elimination of the insulin receptor in fat would by itself lead to changes in metabolism. Indeed, reduced fat is not involved in the mechanism behind the phenomenon of calorie restriction. This can be derived from results obtained by David Harrison (Bar Harbor, ME, USA) and colleagues120, who studied the effects of life-long calorie restriction in genetically obese (ob/ob) and normal mice of the same inbred strain. While the calorie-restricted ob/ob mice still had high levels of adiposity, their maximum lifespan exceeded that of the normal mice and was similar to the lifespan of calorie-restricted normal mice that were much leaner120.
THE LOGIC OF AGING
45
Interestingly, there have been attempts to mimic some of the lifespan-extending mutations in nematodes by targeted inactivation of orthologs of the same genes that conferred longevity in these invertebrate animals. Siegfried Hekimi (Montreal, Canada), who had previously discovered the clk-1 mechanism of lifespan extension in nematodes (see above), inactivated the clk-1 ortholog, mClk1, in the mouse121. It was found that homozygous inactivation of this gene (which is required for the synthesis of coenzyme Q9; see above) in mouse embryonic stem cells yielded cells that are protected from oxidative stress and contain lower levels of spontaneous DNA damage. Whereas the complete mClk1 knockout is embryonically lethal, the heterozygous mice showed an increased lifespan. Interestingly, these investigators observed the loss of the remaining functioning allele of mClk1 during aging in a subset of liver cells, as a consequence of so-called loss of heterozygosity; that is, through the loss of the chromosome or section of the chromosome (see Chapter 4).
2.3.5 INTERRUPTING AGING OR BACK TO NORMAL? The current wealth of mutants of extended lifespan in various organisms appears to confirm that, as predicted by the antagonistic pleiotropy theory, there are genetic pathways that can be interrupted to attenuate aging. It would be logical to expect that such mutations would only lengthen life at the cost of some selective disadvantage, which may or may not be obvious under laboratory conditions. Typically, mutations conferring longevity result in decreased energy metabolism, growth, physical activity, and/or earlylife fecundity. It is likely that such characteristics are not usually the preferred ones in nature with its high external mortality. For the dwarf mice, fitness costs are readily apparent in the form of infertility and hypothyroidism. Weak or sick nematodes or flies may not always be so easily recognizable, but as demonstrated by Gordon Lithgow (Novato, CA, USA) and co-workers, even under laboratory conditions, partial loss of function of DAF-2 results in dramatically reduced fitness as compared to wild-type worms122. However, a number of mutants have been described for which a price to be paid for longevity was not immediately obvious. For such mutants it is conceivable that in the wild there would be a strong selection in favor of the wild-type allele. It is now clear that this may indeed be the case. For example, the age-1 mutation in nematodes, which affects the same pathway as daf-2, shows no obvious effects on fitness under standard laboratory conditions. However, the Lithgow lab exposed mixtures of age-1 mutant and wild-type worms to more natural conditions: cycles of starvation, which are quite common in the wild. A large reduction in the frequency of the mutant allele suggested a substantial difference in relative fitness, which explains the selection of the wild-type allele123. The strong effect of laboratory conditions on the uncovering of aging genes naturally begs the question of whether such ‘laboratory’ genes reflect genuine pro-aging pathways, acting under natural conditions, or are merely artifacts. It is far from sure that loci with
46
THE LOGIC OF AGING
major effects on longevity in laboratory strains of various species will show segregating allelic variation in natural populations. Indeed, even if we adopt the premise that the basic mechanisms that control longevity and determine the rate of aging are common to all multicellular organisms, it is possible that the observed effects on longevity of candidate gene variants will prove to be exclusively associated with laboratory strains and not found in the same animals in the wild. Common laboratory situations generally select for high early fecundity and shorten the lifespan of the organisms under study. For example, Richard Miller (Ann Arbor, MI, USA) and Steve Austad (San Antonio, TX, USA) compared wild-derived mice with laboratory mice of a mixed genetic background and found significantly higher mean and maximum lifespans in the wild mice. Therefore, studies that use laboratory organisms to identify aging-related pathways might identify genes that simply restore the organism’s original lifespan. Such results may not be fully relevant to wild populations124,125. In this respect, even the benefits of calorie restriction—still the only intervention that seems to increase longevity in a large number of species—may be exclusively associated with laboratory animals, which in the wild would forage continuously and endure cycles of starvation. It is conceivable that calorie restriction is the norm in wild populations of mice126. The only valid strategy to test the possibility that genetic variation at candidate loci contributes naturally to phenotypic variance for longevity and aging phenotypes is to demonstrate such associations in the wild (see also Chapter 7). Nevertheless, the discovery of pathways that appear to control aging rate through a synchronized retardation of multiple aging processes is obviously of extreme importance. It suggests that the underlying causes of aging may not be as diverse as often suspected.
2.3.6 HOW DO AGING GENES CAUSE AGING? How could mutations that lead to mild inhibition of growth, reproduction and energy metabolism readily retard aging? We have already seen that many of these aging genes direct trade-offs, which greatly depend on the environmental conditions. Based on the available evidence, a logical explanation would be that the activities of such pro-aging pathways are associated with either an increased generation of somatic damage or a reduced investment in somatic repair and maintenance. In keeping with the theory of antagonistic pleiotropy, it is easy to see why gene variants promoting growth and reproduction are generally favored by evolution, even if their activity would lead to a more rapid rate of somatic damage and a shorter lifespan. Under optimal conditions, investing more in growth and reproduction is a sure bet, taking the ordinary environmental risks into consideration. However, there are situations where reducing growth and reproduction does offer some selective advantage. For example, under certain conditions, such as famine, it would help the organism to shut off reproduction and put all the resources in somatic maintenance and survival. This is exactly what happens when nematodes enter
THE LOGIC OF AGING
47
the so-called dauer stage, a growth-arrested, stress-resistant stage, analogous to spore formation in bacteria or protozoa or hibernation in vertebrates. Such shifts in allocation of resources from growth and reproduction to somatic maintenance and repair are predicted by the disposable soma theory (see above). Are pathways related to growth and reproduction the only candidate aging pathways? Whereas it is certainly possible that all aging pathways may fall into these functional categories, at this stage it is impossible to rule out the existence of other, functionally unrelated pathways. Unfortunately, we will never know because reducing beneficial activities of pathways at early age might be lethal or cause major developmental defects. However, an alternative to genetic screening in finding aging genes and pathways has now emerged in the form of various computational approaches. As we have seen in Chapter 1, genes and proteins exert their physiological functions as networks of interaction rather than individually. In protein–protein interaction networks, or interactome networks, most proteins interact with few partners, whereas a small but significant proportion of proteins, the so-called hubs, interact with many partners. Such networks are called scale-free, which is the norm in cellular networks. Scale-free networks are particularly resistant to random node removal but are extremely sensitive to the targeted removal of hubs127. Daniel Promislow (Athens, GA, USA) compared patterns of connectivity for subsets of yeast proteins associated with senescence. He found that proteins associated with aging have significantly higher connectivity than expected by chance, even when controlling for other factors also associated with connectivity, such as localization of protein expression within the cell128. Similar observations were made by others129 and are consistent with the antagonistic pleiotropy theory for the evolution of senescence. Importantly, they offer an in silico tool to discover new aging genes. However, it should be realized that such in silico approaches are still in their infancy. For example, also on the basis of protein–protein interactions in yeast, it has been suggested that the most highly connected proteins are essential genes130. However, recent results on such networks in mammals (assembled on the basis of interaction data derived from the literature) suggested no such correlation between highly connected genes and lethality of these genes when ablated in either mouse or yeast131. As more complete protein network data are now becoming available for multiple organisms these issues will be resolved and it is certainly conceivable that future in silico searches will either confirm that all aging genes are involved in some aspect of growth, reproduction, and energy metabolism or uncover other, unrelated pleiotropic genes.
2.4 Longevity-assurance genes If aging is the result of the harmful side effects of genes selected for advantages they offer during youth, such effects can be suppressed by the action of other genes, which we can
48
THE LOGIC OF AGING
call longevity genes or longevity-assurance genes. As mentioned above, it is conceivable that the emergence of aging genes and longevity genes goes hand in hand, under the influence of environmental constraints. Longevity genes promote or ensure organismal survival without necessarily playing a direct role in development or maturation. They may encode components of stress-response systems that have apparently evolved to postpone adverse effects associated with endogenous or environmental sources of macromolecular damage. Longevity genes may be highly conserved since endogenous and environmental damage is ubiquitous in all living systems. In particular, DNA-repair systems are ancient, since from the first replicators all living organisms had to cope with agents damaging their nucleic acids (see below and Chapter 4). As already mentioned, regulation of stress-response mechanisms is almost universally associated with increased lifespan. In the nematode, it has been demonstrated that the upregulation of DAF-16 in insulin/IGF-1 mutants affects expression of genes that increase resistance to environmental stress, including the ability to detoxify ROS, which could be the mechanism underlying the increased longevity of these mutants. Indeed, of the more than 40 single-gene mutants in C. elegans that display increased longevity, all increase the ability of the worm to respond to certain types of stress, for example heat, ultraviolet (UV) radiation, and ROS132. One longevity gene, which encodes the OLD-1 transmembrane tyrosine kinase, is stress-inducible and increases longevity by its overexpression. The expression of OLD-1 appears to be dependent on DAF-16 and is required for the lifespan extension of age-1 and daf-2 mutants133. Among various stress responses the heat-shock response has received special attention in view of the capacity of heat-shock proteins to protect cells against protein aggregation. Protein aggregation is caused by disruption of protein-folding homeostasis, which in turn has several possible causes, including oxidative damage, heat, and some forms of genome instability. Heat-shock genes encode chaperones to prevent this and other types of adverse effect (see Chapter 4 for a more detailed discussion of chaperones). Extra copies of the gene encoding the heat-shock protein HSP-16 conferred stress resistance and longevity in the nematode134. Also in this case, the DAF-16 transcription factor was essential for lifespan extension conferred by hsp-16. Overexpression of heat-shock factor (HSF-1), a transcriptional regulator of heatshock genes, increased longevity by about 20%135. Inactivation of hsf-1, but also inactivation of daf-16, accelerated the aggregation of polyglutamine expansion proteins136. As will be discussed in more detail in the next chapter, such aggregation is often caused by the amplification of small triplet repeats in the DNA, a form of genomic instability. Hence, these results suggest that DAF-16 and HSF-1 both act to prevent or attenuate the effects of such genome-instability events. Although daf-16 was required for hsf-1 overexpression to extend lifespan, HSF-1 can function independently of DAF-16136. What this tells us is that upregulation of longevity genes, such as daf-16 or hsf-1, per se is sufficient to increase lifespan. This does not rule out the possibility that the targets of some of these genes are to
THE LOGIC OF AGING
49
be found in metabolism with the explicit purpose of reducing molecular damage at the source rather than fixing its consequences. In addition Drosophila longevity mutants, such as Methuselah, were found to display enhanced resistance to heat, oxidants, and starvation97. In the fly, increased expression of genes involved in protein repair—those encoding protein carboxymethyltransferase and methionine sulfoxide reductase—increase lifespan137,138. Heightened activity of the antioxidant enzymes CuZn- and Mn-superoxide dismutase has also been demonstrated to increase longevity in this organism139,140. Hence, in both nematodes and flies, increased lifespan is caused by increased resistance to stress, either orchestrated in response to reduced growth and energy metabolism, or engineered by manipulating the longevity genes themselves. In the mouse there is also convincing evidence that both the dwarf longevity mutants and mice subjected to calorie restriction display increased resistance to stress, including antioxidant defense, DNA repair, and heat shock141,142. Interestingly, results from Richard Miller and co-workers143 indicate that the increased resistance to stress in the dwarf mice can be demonstrated at the level of skin fibroblasts in culture. Such cells are resistant to multiple forms of cellular stress, including UV light, heat, paraquat, H2O2, and cadmium143,144. Recent results from my own laboratory indicate that dwarf mice show a significantly lower rate of mutation accumulation with age. To detect spontaneous mutations in different organs and tissues of these mice, Ana Maria Garcia, a postdoctoral researcher in the laboratory, crossed a transgenic mouse harboring a lacZ-plasmid reporter construct in its germ line into the homozygous Ames dwarf background. By recovering the lacZ plasmids from genomic DNA of the hybrids and their wild-type littermate controls she was able to compare the mutation frequencies at the reporter locus (for a description of this system see Chapter 6). The results indicate a significantly lower mutation frequency in dwarf mice than in controls. She also subjected these mice to caloric restriction and demonstrated that this also reduced spontaneous mutation frequency, albeit to a lesser extent than dwarfism (Fig. 2.4). These results are in keeping with the delayed occurrence of total neoplastic lesions observed in Ames dwarf mice, which is a major contributing factor to the extended lifespan of these animals113. One reason why mutations conferring lifespan extension based on the inhibition of growth and reproduction show up so readily in a genetic screen could be the selective advantage of mechanisms to delay reproduction and increase survival, as originally proposed by Harrison and Archer145 and Holliday146 to explain the phenomenon of caloric restriction. While the possibility to temporarily halt or attenuate growth and reproduction would be advantageous, actual mutants with continuously downregulated insulin signaling would not survive for many generations, as we have seen from the results obtained by Lithgow (see above). What could be the mechanism underlying the upregulation of multiple longevity genes as a consequence of reducing growth and metabolism? To bridge the gap between metabolism and somatic maintenance, Leonard Guarente
50
THE LOGIC OF AGING 12
7 months
15 months
Mutation frequency (105)
Genotype: P 0.0004 10
Diet: P 0.0040
8 6 4 2 0 CR df
Ad lib df
CR WT
Ad lib WT
Fig. 2.4 Ames dwarf mice (df) and calorically restricted (CR) mice, both long-lived, display a lower spontaneous mutation frequency in different tissues compared to their littermate controls (WT, wild type). Here, only the data for the kidney are shown. The effect of dwarfism is stronger than that of caloric restriction. For details about the measurement of spontaneous mutation frequencies in these mice, see Chapter 6. (A. Garcia, submitted for publication.)
(Cambridge, MA, USA) proposed another master regulator, this time the silencing information regulator 2 (Sir2). Sir2 is the founding member of a phylogenetically conserved family of nicotinamide-adenine dinucleotide (NAD)-dependent histone deacetylases (HDACs), called sirtuins147. Sir2 is required for transcriptional silencing—transcriptional inactivation by altering chromatin structure through deacetylation of histones (Chapter 3)—at the ribosomal DNA (rDNA; the genes coding for rRNA), telomeres (the physical ends of linear eukaryotic chromosomes; Chapter 3), and mating-type loci (regions that differ in DNA sequence between cells of opposite mating-type) of yeast148. The specialized chromatin structure associated with transcriptional silencing is also repressive to recombination. Indeed, Sir2 also suppresses intrachromosomal recombination within the rRNA gene arrays, a process demonstrated by Sinclair and Guarente to lead to an accumulation of rDNA repeats in the form of extrachromosomal circles (ERCs)149. ERCs are self-replicating and are preferentially retained in the mother cells, eventually causing aging by limiting yeast replicative lifespan. Increased Sir2 dose extends yeast replicative lifespan by about 40%, whereas loss of function reduces natural longevity150. Other forms of increased genome instability during replicative aging of yeast, unrelated to rDNA loci or Sir2, have been observed. Most notably, evidence has been obtained for so-called loss-of-heterozygosity events by studying the loss of marker loci affecting colony color151. The steep increase of these loss-of-heterozygosity events was explained by an increasingly impaired ability to correctly detect and repair DNA double-strand
THE LOGIC OF AGING
51
breaks (DSBs), a major form of DNA damage in all organisms and one that can be induced by reactive oxygen (Chapter 4). As we will see in Chapter 6, such forms of genome instability are difficult to detect in mammals. On the other hand, ERCs should be easily detectable, but have not been observed in higher eukaryotes. However, extrachromosomal circular DNA is not uncommon in mammalian tissues where it also is a sign of genomic instability152. Apart from yeast, Sir2 orthologs have been implicated in lifespan regulation of C. elegans and mammals. Tissenbaum and Guarente demonstrated that extra doses of Sir2.1 in the nematode extend adult lifespan by 50%153. This extension of life appeared to be dependent on DAF-16, linking Sir2 to the insulin-signaling pathway. Initially it was thought that Sir2 mediated the silencing of genes in the insulin-signaling pathway in the nematodes; that is, upstream of DAF-16. In mammals orthologs of DAF-16 are the four FOXO proteins, a subgroup of the Forkhead family of proteins, transcriptional regulators characterized by a conserved DNA-binding domain termed the forkhead box154 (see Chapter 3). It turned out that FOXO itself is a direct, functional target for Sir2155. In mammalian cells, Sir2 (SIRT1 in humans) can deacetylate FOXO, increasing its ability to induce cell-cycle arrest and resistance to oxidative stress, but inhibiting FOXO’s ability to induce apoptosis156,157. Similarly, SIRT1 can deacetylate p53 and attenuate its transcriptional activity158. Hence, Sir2 seems to specifically modulate activities that contribute to survival. The role of FOXO in increasing cellular defense is underscored by the observation that, in rat fibroblast cells, this transcription factor induces the repair of damaged DNA and upregulates the growth arrest and DNA damage response gene Gadd45a159. How can Sir2 regulate survival in different species? A possible answer can be found in the NAD hydrolysis, an integral step of the deacetylation reaction carried out by Sir2 and other sirtuins, which leads to the consumption of one NAD molecule for each deacetylated lysine residue. It has been shown that the prolongevity effect of caloric restriction— accomplished in yeast by limiting glucose availability—is lost in the absence of Sir2160. This suggests that calorie restriction slows aging by activating Sir2. Guarente et al.160, then, proposed that Sir2 may connect energy metabolism to lifespan through the absolute requirement of Sir2’s activity for NAD. If cellular NAD levels are low, then Sir2’s deacetylase activity could be attenuated and vice versa. As demonstrated by the Guarente laboratory, calorie restriction decreases the levels of NADH, a competitive inhibitor of Sir2. Therefore, an increased NAD/NADH ratio in calorie restriction could underlie the increased Sir2 activity, which in turn would repress recombination at the rDNA locus, thereby slowing the formation of toxic rDNA circles and increase lifespan. This is exactly what has been shown160. However, it is not easy to determine the effective concentrations of NAD and NADH in living cells and the role of physiological variation in NADH in affecting Sir2 activity is controversial161. Also, the role of Sir2 itself as a key factor in caloric restriction has been disputed162.
52
THE LOGIC OF AGING
Interestingly, an increase in Sir2 activity by calorie restriction has been demonstrated in mammalian cells: expression of the ortholog of Sir2, SIRT1, is induced by calorie restriction in rats, as well as in human cells treated with serum from these animals163. It was also shown that SIRT1 suppresses apoptosis, suggesting that the induction of SIRT1 by calorie restriction represents a survival response, similar to its deacetylation of FOXO and p53 as described above. Whether or not Sir2 is a master regulator, as some believe it is, there can be no doubt that intricate mechanisms have evolved to regulate the survival of an organism as part of its life-history strategy. As we have seen, these longevity mechanisms can be upregulated directly or by dampening pro-aging pathways through a reduction of growth and reproductive efforts. A more detailed understanding of the nature of the damage that limits our lifespan, and the evolutionary history of its interaction with the longevity systems that emerged to counter its effects, may give us insight into the causes of aging.
2.5 Somatic damage and the aging genome Somatic damage accumulation appears to be a general characteristic of living organisms. In a broad sense such damage can vary from infectious agents to DNA damage and mutations. Exposure is dependent on a host of variables, including species-specific endogenous and ecological factors. Nevertheless, some universal principles basic to life or at least basic to most eukaryotes suggest similarities in exposure. For example all aerobic organisms are exposed to ROS. Many organisms are exposed to body heat and such environmental agents as UV and ionizing radiation. Such similarities in exposure to damaging agents is in keeping with the universality of pathways of life extension originally uncovered in nematodes and the resistance of virtually all these longevity mutants to ROS, heat, and other damaging agents. The most ancient example of damage accumulation in the living world is damage to nucleic acids, first RNA then DNA. Genetic damage posed both a fundamental problem and an opportunity for living systems. A problem, because genetic damage essentially prevents the perpetuation of life, since it interferes with replication (and transcription); an opportunity, because it allows the generation of genetic variation through errors in replicating a damaged template. Similar to the trade-offs mentioned above between reproduction and somatic maintenance, the relative stability of a genome is optimized to the life history of the organism. The necessity of mutations from the first replicators onwards is a strong argument to consider DNA damage and genetic errors as the original instigators and major drivers of aging in the living world. A second, logical argument to consider the DNA of the genome as the Achilles heel of an aging organism is the irreversibility of unwanted sequence changes in view of the lack
THE LOGIC OF AGING
53
of a back-up template. This is in contrast to proteins, which at least in principle can be easily replaced, with the corresponding gene as the template. Indeed, the maintenance of genomic DNA is of crucial importance to survival because its alteration by mutation is essentially irreversible and has the potential to affect all downstream processes. Third, as explained in Chapter 5, there is now overwhelming evidence that in both humans and mice heritable defects in genes involved in maintaining the integrity of the genome cause symptoms of premature aging. The first replicating nucleic acids, almost certainly RNA, evolved almost 4 billion years ago in an environment with little molecular oxygen, but high fluxes of UV radiation due to the absence of an ozone layer. Based on this constant threat of DNA-damaging agents and the absolute need to increase replication fidelity from the primitive RNA-replicating systems to the much larger RNA and DNA genomes, it is now generally believed that recombinational repair was the first bona fide DNA-repair system that evolved164,165 (see also Chapter 4). Although mutations are essential as the ultimate source of genetic variation upon which evolution depends, too many mutations will reduce fitness and can lead to population extinction. As described above, due to Muller’s ratchet73, asexual lineages will tend to lose mutation-free genomes (due to genetic drift) and inevitably suffer from loss of viability. This is especially relevant for small populations and relatively easy to demonstrate in simple unicellular organisms, such as E. coli. For example, Kibota and Lynch demonstrated that deleterious mutations of small effect escaped selection in E. coli lines with repeated population bottlenecks, resulting in decreased fitness166. Elena and Lensky, also in E. coli, demonstrated that randomly introduced mutations interact to negatively affect fitness; that is, the relationship between mutation number and decreased fitness was nearly log-linear167. Is it realistic to interpret loss of fitness caused by mutation accumulation in unicellular organisms as a parallel to aging of somatic cells of metazoa? As described by Graham Bell (see above), senescence-like phenomena have been observed in asexually propagated protozoa lineages, and were attributed to the accumulation of deleterious mutations69. Bell explained the recombination of genetic material between different lineages—sexual outcrossing—in terms of an exogenous repair mechanism functioning by creating variance on which selection can act effectively to reduce mutational load. However, Bell has argued against a parallel between mutation-induced senescence in protozoa and aging of somatic cells in metazoa. Although he did not rule out a role for mutation accumulation in aging of somatic cells of a mammal, he considered it unsound because senescence will evolve—as a by-product of selection for increased early reproduction—in the soma, but not the germ line of metazoa. In protozoa, germ cells and somatic cells are of course identical, but there is no a priori reason why Muller’s ratchet would not apply to somatic cell lineages as well. Mutations have been with us since the origin of life and were the primary condition for the evolution of the wide variety of species on our planet. Indeed, organisms with
54
THE LOGIC OF AGING
sufficient tolerance of their genome maintenance systems to allow for a certain level of mutations have the best chance to survive environmental challenges. Visible examples of this strategy are various infectious diseases in which the bacteria or viruses have the capacity to evade drugs and the immune system by rapidly altering their antibiotic resistance or cellular receptor repertoire. Indeed, pathogenic species, such as Streptococcus pneumoniae, harbor more plastic genomes with more potential for host adaptation than bacteria adapted to a non-threatened lifestyle, such as Streptococcus thermophilus, used for the manufacture of yogurt and cheese168. In this particular case the difference is due to the lack in the deadly pneumococcus of two enzymes involved in the repair of DNA DSBs (see Chapter 4). This situation is very similar to cancer. Tumor cells often gain the capability, like pathogenic microorganisms, to rapidly alter their genotypes, thereby creating new attributes to grow more efficiently, metastasize, and evade both the immune system and various drug treatments. While in microorganisms increased genetic variation is facilitated by stress-induced mutagenesis169, in human cancers mutator phenotypes caused by the inactivation of genome-maintenance systems temporarily result in an increased opportunity to adapt to adverse conditions170. Whereas mutations in combination with recombination provide innovation at the population level, they can cause individual failure and decreased fitness. Organisms have several ways to protect themselves against mutations or their consequences. There is first of all a host of genome-maintenance systems acting to prevent mutations by efficiently, and most of all accurately, repairing chemical DNA damage. However, as we will see in Chapter 4, whereas these systems are able to repair most if not all of the various types of DNA damage continuously induced in living cells, they are imperfect. Indeed, apart from DNA-synthesis errors per se, virtually all mutations are due to so-called error-prone repair; that is, mistakes made during the repair of such chemical lesions. A less obvious defense against mutations and their consequences is the fault-tolerant way genetic information is encoded; that is, in a redundant manner with multiple copies of a gene or with different genes or pathways that can carry out similar functions. A third way to limit the impact of mutations is by avoiding extreme optimization of function, which implies tolerance for defects. Finally, organisms use so-called genetic capacitance, which involves proteins that buffer the effects of mutations, such as the heat-shock proteins or molecular chaperones that were demonstrated to increase nematode lifespan (see above). Overall, therefore, organisms are robust: they are able to maintain function against various perturbations. Nevertheless, even robustness that is ubiquitous in biological systems has inherent trade-offs and failure patterns. These become especially apparent in multicellular organisms. While unicellular organisms can out-select mutation accumulation, in multicellular organisms this is only true for the germ line. Indeed, in animals germ-line cells are set aside very early in life, after which there is ample opportunity for selection against gametes and gamete combinations carrying deleterious mutations. (Plants do not set aside a germ-line early on in life, but instead rely on stringent selection
THE LOGIC OF AGING
55
during the haploid gametophytic phase and selection during somatic growth when cell lineages carrying deleterious mutations are impaired in growth and development and therefore less likely to contribute to gamete formation171.) Even then, signs of system failure are evident even very early in life. In humans, for example, this translates in the relatively high percentage of spontaneous abortions and stillbirths, which are most likely caused by mutations172 (see Chapter 6). It is estimated that every human newborn has acquired as many as 100 new mutations per genome, three of which might be deleterious173. In the somatic genome of multicellular organisms selection against deleterious mutations is more difficult than in the germ line. Obviously, cell loss or cellular loss of function due to deleterious mutations will contribute to functional decline of the organ and may cause disease, most notably cancer. In theory, optimization of genome-maintenance mechanisms for only the somatic cells could prevent any significant mutation accumulation. However, evolutionary theory would not predict a maximization of cellular maintenance and repair in view of the gradual decline of the force of natural selection after the age of first reproduction (see above). In most cases, our fault-tolerant genomes would allow mutation accumulation well into the reproductive period without appreciable loss of function. However, it is unlikely that even slight increases in mutation loads thereafter would have no consequences. An observation that has often been used to discard the idea that mutation accumulation could adversely affect physiological functioning and cause aging and death is the identical lifespans of haploid and diploid wasps of the genus Habrobracon174. As in other hymenopterans, in Habrobracon unfertilized eggs become haploid males and fertilized eggs that are homozygous and heterozygous at the sex locus develop into diploid males and females, respectively. While X-irradiation of larvae, pupae, or adults reduced lifespan of the haploid variant more than that of its diploid counterpart, their normal aging rate was the same. Somewhat surprisingly, the results of this isolated observation, which has not been confirmed in other species for which both haploid and diploid variants exist, has been widely interpreted to discard the somatic mutation theory of aging. As argued by Alec Morley (Adelaide, Australia) in a lucid paper not cited frequently enough, the Habrobracon results, even if reproducibly obtained in multiple species, do not exclude a causal role in aging of dominant or co-dominant mutations, which may be more important in terms of functional significance upon their random induction in somatic cells175 (see also Chapter 6). Most importantly, although the authors assumed that mutations were the cause of the reduced lifespan in the wasps after X-irradiation, this is unlikely to be true. Indeed, adverse effects after an acute dose of X-rays are more likely to be caused by the toxic effects of DNA damage rather than mutations. As we shall see in Chapters 4 and 6, mutations are the result, mainly, of erroneous processing of DNA damage; their introduction requires time. Acute effects of DNA damage are likely to be mitigated by an additional copy of each gene as in the diploid situation. The absence of a
56
THE LOGIC OF AGING Growth and reproduction
Increased longevity
Somatic maintenance
Somatic damage
ROS
GH p66SHC
Sir2
IGF-1 ROS
p53
IGF-1R Apoptosis PI3K
Akt
Aging
FOXO3
Antioxidant defense DNA repair Stress resistance Longevity
Fig. 2.5 Some of the factors involved in the genetic control of longevity and aging. Black represents pro-aging gene products for which reduced function or loss of function increases lifespan. In white are the pro-longevity gene products, increased expression of which extends lifespan (and vice versa). Arrows indicate agonists, and bars indicate antagonists. See text for further details. IGF-1R, IGF-1 receptor. Re-drawn with permission from ref. 443.
difference in lifespan between the unirradiated haploid and diploid animals is not inconsistent with a causal role of dominant mutations in aging. Indeed, as will be argued later (Chapter 7), mutations in the form of large rearrangements could easily lead to genome destabilization and an aberrant pattern of gene expression. This would be the case for both haploid and diploid genomes. An in-depth investigation of the haploid versus diploid aging wasps would be necessary to test this hypothesis. In summary, from the earliest replicators onwards the logic of life with its need for evolvability has dictated the logic of aging. Species-specific lifespans reflect the balance between growth and reproduction and somatic maintenance and repair (Fig. 2.5). Somatic maintenance and repair initially only involved the need to preserve the genome, but later also other cellular and organismal structures became important. It is possible that even with the emergence of billions of different species of increasing complexity this basic principle is still valid and the genome may still be the most relevant target of the aging process. What has changed, however, is the complexity of genome structure and function, and with it the diversity of threats to its integrity. This will be discussed in the next chapter.
3
Genome structure and function
As described in Chapter 1, the science of aging has evolved in concert with an emerging insight into both the logic of life and its structure–function relationships. Insight into the logic of life gave us insight into the logic of aging, as outlined in Chapter 2. However, whereas we know why we age, the mechanistic basis of the process as it takes place in different species is still unclear. Full knowledge of how we age can only emerge hand in hand with the further uncovering of the basic principles underlying the nature of the living world. In the life sciences it is now slowly being realized that the unraveling of how life is ordered cannot be separated from enquiries into what causes its progressive disorder and ultimate demise. Genome structure and function are central to the organization of life. They determine our species-specific characteristics and provide the conditions for individual development and maturation. As such, the genome may bear the roots of its own destruction and with it the causes of age-related cellular degeneration and death. The genome, a term coined in 1920 by Hans Winkler, a Professor of Botany at the University of Hamburg, to designate the haploid chromosome set, has more recently been considered to mean the whole complement of genes of an organism and sometimes also the sum total of coding and non-coding DNA sequences (see also Chapter 1). Here I will define a genome in somewhat broader terms as the organelle or physical entity that carries out all genetic transactions in a cell or organism. This necessitates a three-dimensional view of the genome as a coordinated ensemble of gene action in the context of a series of structure–function relationships, which ultimately represents the complete network of processes that defines a living organism. Such a genome interacts extensively with other molecules. To understand how the genome exerts its function as the ultimate determinant of both order and progressive disorder it is necessary to know its structure and functional organization. There are recent textbooks and extensive review articles to which I refer for comprehensive information on various aspects of genome structure and function176,177. Here, I will focus predominantly on those components of genome structure that have the potential to drive its demise in somatic cells over time, based on our current knowledge of the sources of somatic damage in aging organisms and mechanisms of genomic instability. To this end, I will address genome structure and organization at three levels: first, DNA primary structure and sequence; second, higher-order DNA structure; and third, nuclear architecture. In all cases the focus will be on the impact of these structural
58
GENOME STRUCTURE AND FUNCTION
levels on cellular function, especially patterns of regulation of gene transcription, which will be discussed in the last section of this chapter.
3.1 DNA primary structure As discussed in Chapter 1, the landmark discovery of the double-helical structure of DNA by Watson and Crick in 1953 provided a logical basis for the role of the genome in both the perpetuation of life in evolutionary time and its functional organization (see Fig. 1.4). The DNA of the genome is a polymer consisting of four different types of monomer units, termed nucleotides. Each nucleotide consists of a five-carbon sugar (deoxyribose), a nitrogen-containing base attached to the sugar, and a phosphate group. The differences between the four nucleotides are determined by the different bases: adenine (A), guanine (G), cytosine (C), and thymine (T). Adenine and guanine are the purine bases and the larger of the two types, with cytosine and thymine the relatively small pyrimidine bases. RNA is different from DNA, since its nucleotides contain a ribose instead of a deoxyribose as the sugar, and uracil (U) replaces the thymine base. Apart from their major—amino and keto—tautomeric forms, the DNA bases can adopt alternate forms; the imino form for adenine and cytosine and the enol form for guanine and thymine. Although these structural isomers are rare, they do alter base-pairing properties, which makes them a source of mutations when accidentally present during DNA replicative or repair synthesis. The monomer units in nucleic acids are connected through the phosphate residue attached to the hydroxyl group on the 5' carbon of one unit and the 3' hydroxyl on the next one (Fig. 3.1). This forms a phosphodiester link between two residues (shown as p; e.g. CpG), which can lead to very long nucleic acids—up to billions of units. The heteropolymeric character of nucleic acids is the basis for their role in information storage and transmission in the form of a base sequence code of nucleotides. As discovered by Watson and Crick12, the DNA of the genome is a double-stranded macromolecule, with a double-helical structure of two polynucleotide chains, held together by specific pairing of AT and GC base pairs stacked on one another with their planes nearly perpendicular to the helix axis (Fig. 3.2). This configuration allows strong van-der-Waals interactions between the bases. DNA has both a sense of direction and individuality. The phosphodiester linkage between monomer units is always between the 5' carbon of one monomer and the 3' carbon of the next. The two strands in the double-helical model run in opposite directions (Fig. 3.2). Individuality is, of course, determined by the sequence of its bases; that is, the nucleotide sequence. It is in the primary structure of a nucleic acid that genetic
GENOME STRUCTURE AND FUNCTION (a)
O
Thymine (T) H
H3C O2 O
P
59
N
H
O
N
CH2
O
O O
2
H
Phosphate
N
H
H
Sugar
N
H
N
H
O
N O
P
Adenine (A)
H
H
H
O
N
CH2 O
O–
H
H
H
H
Cytosine (C)
H
H
N
H
H
H
H
N
O H O
P
O
O
N
CH2 O
Guanine (G)
O2
H
H
H
O
H H
N P
N
H
O O
H
N
O
CH2 O
O2
H
H N
N H
H H
H OH
H
Fig. 3.1 Primary structure of DNA.
information is stored as a four-letter code, encoding genes and gene-regulatory or structural sequences. Polynucleotides are thermodynamically unstable in vivo but their hydrolysis is exceedingly slow unless catalyzed. Similarly, whereas dehydration would readily allow adding nucleotide residues to a nucleic acid chain, the thermodynamics of this reaction are unfavorable and require high-energy nucleoside triphosphates (ATPs). Hence, both the breakdown and synthesis of nucleic acids require enzymatic processes; in their absence DNA is sufficiently stable to serve as a useful repository of genetic information. Nevertheless, DNA is not completely stable under physiological conditions. Apart from errors during its replication, DNA can be damaged through its interaction with a variety of reactive chemicals
60
GENOME STRUCTURE AND FUNCTION (b)
5
3
Base pairs
Phosphates P P
Deoxyribose
P P
P P
P P
P P
Major groove
Minor groove
3
5
Fig. 3.2 The Watson and Crick model of the DNA double helix.
or radiation. Even in the absence of environmental challenges, the DNA in each cell of an organism suffers from a multitude of endogenous damage, resulting from spontaneous hydrolysis and oxidation (in aerobic organisms). Spontaneous DNA damage, DNA repair, and genome instability are discussed extensively in Chapters 4 and 6. Nucleic acids can have several secondary structures, including the right-handed A and B helices, which differ mainly in the position of the bases with respect to the helix axis, and the left-handed Z helix, discovered by Alexander Rich (Cambridge, MA, USA) and co-workers in 1979178. Probably, most of the DNA in the aqueous milieu of cells in vivo is in the B form, whereas the A conformation is adopted by double-stranded RNA and DNA–RNA hybrids. The surface of the B helix contains two different grooves, called the major and minor groove (Fig. 3.2). Proteins that interact with DNA often make contact in these grooves, especially the major groove, which is larger and more suitable for sequence-specific recognition. Unfortunately, since these areas contain an abundance of reactive sites they are also often the target of DNA-damaging agents (see Chapter 6). Z-DNA occurs in regions of polynucleotides having alternating purine–pyrimidine sequences with a substantial fraction of the C residues methylated to form 5-methylcytosine.
GENOME STRUCTURE AND FUNCTION
61
5 5 3
5 X junction or cruciform
G quadruplex
Hairpin junction
Fig. 3.3 Examples of multi-stranded DNA structures as they can occur in vivo. In the G quadruplex the four DNA strands can have different polarities, from all strands parallel to alternating antiparallel strands.
This modification of C occurs to a significant extent in natural DNA, where it plays a role in such transactions as the regulation of gene transcription (see below). Apart from double-helical DNA, the genome in vivo may also contain multistranded DNA structures such as triplexes, quadruplexes, and junctions (Fig. 3.3). Some of these, for example Holliday junctions, can be intermediates of DNA transactions, such as homologous recombination (Chapter 4). A large variety of others may form spontaneously in an aqueous milieu and could have important biological implications. Examples of genomic DNA sequences with the potential to form such structures are the telomeric regions at the end of chromosomes and the so-called promoter regions that regulate gene action (see below). Most of these sequences have continuous stretches of guanine nucleotides, and the ability of the guanine base to form tetrads (DNA tetraplexes or G quadruplexes), with four guanines in a plane, lies at the crux of these complex structures. Among these, the structures formed by the telomere sequences have been the most widely investigated in relation to aging, especially in relation to the gene defects causing Werner syndrome, a segmental progeroid syndrome due to the inactivation of the WRN gene, a RecQ helicase thought to be involved in resolving such structures (Chapters 4 and 6).
3.1.1 GENES IN GENOMES As we have seen in Chapter 1, the genes, hidden in the primary structure of the DNA of the genome, give rise to phenotypes through the generation of corresponding protein products by transcription and translation. However, the relationship between genes as the units of inheritance and their physical structure has become somewhat nebulous in view of the recognized need for accessory DNA sequences to regulate transcription. Indeed, genomes are complex and represent more than a bag of genes. This is illustrated by the results of the comparative analysis of the hundreds of prokaryotic and eukaryotic genomes that have now been decoded, including the complete sequences of five
62
GENOME STRUCTURE AND FUNCTION
Table 3.1 Genome features of several organisms Organism Homo sapiens Pan troglodytes Mus musculus Caenorhabditis elegans Drosophila melanogaster Saccharomyces cerevisiae Escherichia coli
Approx. genome size (Mb)a
Diploid no. of chromosomes
Approx. no. of genesb
Human gene homologs (%)c
3400 3690 3450 100 180 12 4.6
46 48 40 12 8 32 1
30 000 30 000 30 000 19 100 13 600 6300 3200
– 87 79 31 39 11 2
a
Source: www.genomesize.com Source: www.ornl.gov/sci/techresources/Human_Genome/faq/compgen.shtml c Percentage of human genes with homologs in the organism of interest. Source: http://eugenes.org/all/hgsummary.html
b
mammals: human, mouse, rat, dog, and cow. Genomes can vary in size from about 0.5 million bp (0.5 megabases or Mb) in Nanoarchaeum equitans, the smallest genome of a true organism yet found, with slightly more than 500 genes, to 3000 Mb in the human genome, with about 30 000 genes (see www.ncbi.nlm.nih.gov/genomes/ for all genomerelated databases). In general, such comparative analysis has revealed increases in complexity from prokaryotes to multicellular eukaryotes. Whereas genome size appears to be generally associated with organismal complexity this is much less obvious for the number of genes. For example, whereas humans have only about 10 000 more genes than C. elegans, the human genome is 30-fold larger than that of the nematode (see Table 3.1 for a comparison of genome features among selected species). Virtually all of the increase in genome size from prokaryotes to mammals is caused by the addition of non-genic DNA. Whereas E. coli has about 4000 protein-coding genes comprising almost 90% of its total sequence, in the human (or mouse) genome only slightly more than 1% is protein-coding sequence. Whereas there is an ongoing debate about the possible function of much of this non-genic DNA, one explanation that has been put forward for the origin of genome complexity in vertebrates is the enormous reduction in population size associated with the emergence of higher eukaryotes of larger size. This would permit an initially non-adaptive restructuring of eukaryotic genomes by genetic drift, providing novel substrates for the secondary evolution of phenotypic complexity179. However, it should be realized that genome size varies enormously among organisms without any obvious relationship to complexity. For example, Amoeba dubia, a small protozoan species of indisputably lower complexity than Homo sapiens, has a genome size of 670 billion bp, more than 100-fold the size of the human genome. Before drawing the conclusion that these figures demonstrate that large genomes and complexity are not associated it should be realized that these extraordinary large genome sizes are not based on complete sequence information but on biochemical measurements. In such cases genome size may be greatly exaggerated since they do not account for
GENOME STRUCTURE AND FUNCTION
63
Protein-coding 1% Introns 24%
Mobile genetic elements 45%
Other 30%
Fig. 3.4 Organization of the DNA sequence content of the genome. Other types of sequence, apart from mobile genetic elements and genes, are the various families of tandem repeats, noncoding RNAs, pseudogenes, and some as yet still unidentified DNA.
mitochondrial DNA (mtDNA), the amount of which can be substantial, and a sometimes high levels of polyploidy (see the Animal Genome Size Database, www.genomesize.com). Non-genic DNA in the mammalian genome includes the relatively small 5' and 3' untranslated regions flanking the genes, spliceosomal introns, dispersed within genes and comprising as much as 95% of the genes, mobile and repetitive genetic elements, tandem repeats at centromeres and telomeres, and non-coding RNA genes (Fig. 3.4). Whereas in the past much of this type of DNA has been considered as junk DNA (because there seemed to be no obvious benefit to the host) or selfish DNA (because its only function seems to be to make more copies of itself), it is now clear that at least a sizable part of it must be functional180. This is suggested by the evolutionary conservation of many nongenic DNA sequences. As explained in Chapter 1 once complete genome sequences were available it became possible to compare them in their entirety (e.g. human and mouse) for sequences of high similarity as a result of negative selection. Inter-species studies of homology revealed that a large part of the human or mouse genome—as much as about 20%—is subject to evolutionary constraint180. It should be noted that the situation for families of repeat elements (discussed below) has its own peculiarities. Indeed, such families, which are abundantly present in mammalian genomes, can be highly divergent between species. However, there is a high level of sequence conservation between repeats, which can include genes such as rRNA genes, within a species. This is called concerted evolution and based on what has been termed genetic drive181. At least some repeat families, certainly the satellites found at centromeres (see below) have a highly conserved biological function (in spite of their high divergence between species).
64
GENOME STRUCTURE AND FUNCTION
Interestingly, whereas high levels of conservation were found (as expected) in proteincoding parts of genes, 1–2% of the human genome represents equally highly conserved, non-genic sequences. The distribution of such sequences is negatively correlated with the distribution of genes182. This suggests that their function either involves long-distance regulation of gene expression or some other important role, such as in chromosomal transactions or in the interaction of the genome with other cellular and nuclear structures (see below). Another characteristic feature of complex genomes, such as the human genome, is that while only about 1% may encode proteins, a much larger part (perhaps almost 50%180) is transcribed. It is possible that at least a part of these non-coding RNA molecules function in the regulation of gene transcription. For one category of non-coding RNAs (see below), the so-called miRNAs, this has now been demonstrated. What this picture of the genome suggests is that the structural and physiological complexity of organisms is dependent not so much on their genes, but on the way the activity of these genes is regulated. This can be very different from species to species even if most of their protein sequences and core regulatory regions are virtually identical. Indeed, the regulation of gene expression takes place at many levels, of which the relative position of genes and regulatory regions is of critical importance. As noticed by the late Alan Wilson (1934–1991), among placental mammals the variation in karyotype (chromosome number, large inversions or deletions) is much more dramatic than in the primary DNA sequence183. Indeed, based on early cytogenetic comparisons one would never believe that the mammalian genome is so highly conserved at the sequence level. During evolution of the different mammalian lineages, many genome rearrangements have scrambled the relative positions of genes and other homologous sequences beyond recognition. As noticed by several authors, including those active in the science of aging184, it is unlikely that the considerable differences between human and chimpanzee, not least the approximately 2-fold difference in maximum lifespan, are due to differences in their protein sequences. Instead, there is ample reason to believe that much of the phenotypic difference between closely related species is explained by evolutionary changes in gene regulation. Apparently, a large number of genome rearrangements can occur over a relatively short evolutionary time span.
3.1.2 GENE DISPERSION AND MODULAR ORGANIZATION In the mammalian genome, genes are clustered densely into small islands in desert-like expanses containing few or no genes. While their protein-coding sequences occupy only about 1% of the entire DNA sequence, their interruption by introns allows them to be spread over huge distances. Introns, which comprise about 24% of the genomic sequence, are non-coding stretches of DNA that are transcribed, but eventually excised from the
GENOME STRUCTURE AND FUNCTION
65
primary transcript, thereby fusing the remaining parts, termed exons into the final, protein-coding mRNA. Introns almost never appear in prokaryotic cells and are rare in single-celled eukaryotes, but in multicellular animals and plants almost every gene has introns. Both terms—intron and exon—were coined by Walter Gilbert (Cambridge, MA, USA) in 1978, who postulated that exons were originally minigenes corresponding to current protein domains (structurally and functionally defined, semi-independent parts of a protein)185. At a later stage in evolution, the minigenes were assembled to make whole genes, with introns as the functionless pieces that held the exons together. Bacteria may have lost introns in later evolutionary stages. Now we know that exons do not always map to those domains. Indeed, another school of thought argues that introns are a relatively recent arrival in the eukaryotic lineage and necessary to help generate the diversity of regulatory mechanisms that are required to control gene expression in multicellular, highly differentiated organisms186. In this view, prokaryotes do not have introns because they never had them in the first place. At first sight introns seem to impose a selective disadvantage on their host genes by increasing the chance of mutation to yield defective alleles. On the other hand, introns provide for alternative splicing (making different combinations of exons), which is advantageous to eukaryotes because it allows single genes to encode multiple protein isoforms, a major source of genetic diversity. Introns may also have other functions, including a role in nucleosome formation and in anchoring chromatin loops to the nuclear matrix and to chromosome scaffolds (see below). The modular nature of genes in higher eukaryotes with so many introns has led to the substitution of the term transcription unit for the original gene or cistron, which remains valid in prokaryotes. The concept of a gene in mammalian genomes is blurred, not only by alternative splicing, which creates different proteins from one gene, but also by the discovery that different genes can physically overlap and that sometimes an exonic sequence in one gene is part of an intron in another.
3.1.3 MOBILE GENETIC ELEMENTS AND DISPERSED REPEATS Originally discovered by Barbara McClintock (1902–1992) in maize187, mobile genetic elements, such as Alu and L1 sequences, make up about 45% of the human genome. Mobile genetic elements are pieces of DNA that can move from one place in the genome to another. By insertion into a gene, they can be the cause of the mutations responsible for some cases of human genetic diseases, including hemophilia188. Like for introns there has been a debate as to whether such sequences have a function or should be merely considered as junk DNA. There is ample evidence that mobile genetic elements can contribute to genome evolution by providing regulatory elements to neighboring genes or by helping to create new combinations of exons, promoters, and enhancers189. It is also possible
66
GENOME STRUCTURE AND FUNCTION
that these repeat elements have some structural role and help providing the necessary physical organization required for effective, integrated genome functioning. There are four main families of mobile genetic elements in the human genome. First, the long interspersed nuclear elements (LINEs), of more than 6 kb; second, small interspersed nuclear elements (SINEs), of less than 500 bp, such as Alu; third, the retroviruslike transposons; and fourth the DNA transposons. The first three proliferate via an RNA intermediate; only DNA transposons are real jumping genes, moving from one place to another by using a cut-and-paste mechanism. Here I will only briefly discuss the LINE1 element, also termed L1. The L1 retrotransposon is the most abundant dispersed repeat sequence, comprising over one-third of the entire human genome190. This sequence generates a copy of itself, through transcription followed by reverse transcription, which then integrates elsewhere. About 100 L1 elements in the human genome are still active to retrotranspose and are therefore an ongoing source of mutations. Because L1 elements are so widespread and present in so many copies, they are also a target for erroneous homologous recombination (see Chapter 4 for a detailed description). In view of these adverse effects, one might wonder why L1 repeats, in spite of their evolutionary benefits, are tolerated in the genome and have not disappeared. Apart from the fact that the efficacy of natural selection would be insufficient to prevent the spread of this evolutionary useful sequence—provided the activity of this aggressive mutator is dampened sufficiently long to allow its host to reach the reproductive age—L1 repeats are likely to have some benefits. L1 repeats may serve as attenuators of gene-transcriptional activity by spreading heterochromatin formation through cytosine methylation of its promoter191 (see also below). Another suggested function of L1 elements involved their capacity to cause mutations. It has recently been demonstrated that an engineered human L1 element can retrotranspose in neuronal precursors derived from rat hippocampus neural stem cells192. The resulting retrotransposition events sometimes altered the expression of neuronal genes, affecting neuronal differentiation. The same investigators also showed that the L1 element was active in transgenic mice, resulting in neuronal somatic mosaicism. They suggested that L1 repeats could contribute in this way to neuronal somatic diversification in the developing brain, possibly explaining individual differences in brain organization and function. Whether or not this will prove to be correct (see also Chapter 6), these observations underscore the fact that somatic genomes are not static, but diverge during development, and possibly also during aging, because of de novo DNA sequence changes. Totally different types of dispersed repeat are the simple sequence repeat families termed microsatellites and minisatellites. These elements, with a unit size of between about 2 and 20 nucleotides, are organized as loci of tandem repeats dispersed through the genome. They tend to be unstable, with some loci showing dramatically high mutation frequencies, often based on so-called slippage replication errors, leading to copy-number
GENOME STRUCTURE AND FUNCTION
67
Anticipation
VNTR polymorphism (Between chromosomes)
Somatic instability
Fig. 3.5 Variable number of tandem repeat (VNTR) instability mainly resides in micro- and minisatellite loci. Extensive copy-number variation at these loci is present in the germ line, but also occurs spontaneously in somatic cells or can be induced by radiation. Anticipation is the increase in copy number of microsatellite repeats in the germ line, which can suddenly become manifest as a genetic disease. See the text for more details.
variation (Fig. 3.5). It explains why these loci are so polymorphic in the germ line, which makes them highly suitable for DNA-based identity testing (DNA fingerprinting) and as markers in genetic linkage studies aimed at discovering the location of disease genes193. Mini- and microsatellites are unstable in both the germ line (parent-to-offspring) and somatic cells. For microsatellite loci (including tracts of the same base) this may be due mostly to slippage replication errors, whereas minisatellite variation appears to involve mainly unequal exchange during recombination. Indeed, microsatellite instability is especially prominent in the presence of a defect in DNA-mismatch repair, the system that edits the newly synthesized DNA strand to correct mismatches, including those that result from slippage replication errors. Tumors arising in patients with a heritable form of cancer caused by a defect in DNA-mismatch repair often display this form of microsatellite instability since they are unable to repair the mismatches that are a consequence of slippage replication errors (see Chapter 4). Microsatellite instability has also been associated with an intriguing set of heritable diseases, including Huntington’s disease, fragile X syndrome, and myotonic dystrophy. These diseases are caused by an increase in the number of copies of the repeat beyond its normal range of about 6–55 repeats. This so-called genetic anticipation (Fig. 3.5), which occurs in the germ line, is responsible for the increasing severity of an inherited disease during intergenerational transmission. Studies of these diseases and the repeats involved have shown that this type of genomic instability is highly dynamic194. Disease-related instability of the repeats is not limited to the germ line but occurs also in somatic cells in a tissue-dependent manner. We have already seen that in C. elegans both DAF-16 and the regulator of the heat-shock response, HSF-1, protect against the adverse consequences of
68
GENOME STRUCTURE AND FUNCTION
copy-number amplification of microsatellites as occurs in Huntington’s disease; that is, polyglutamine aggregation136 (Chapter 2). The possible contribution to aging of this form of genome instability will be discussed in Chapter 6.
3.1.4 TANDEM REPEATS AND CHROMOSOME STRUCTURE Apart from the dispersed loci of short tandem repeats, most eukaryotes have many copies of tandemly repeated DNA sequences located around their centromeres and at the telomeres. Centromeres (the points at which spindle microtubules attach; see below) are responsible for chromosome segregation during mitosis and meiosis and contain mainly so-called -satellite repeats. These highly homogeneous repeats, which are not transcribed, have a basic 170-bp unit and are required for centromere function in a way that is still unclear195. As mentioned above, these sequences are subject to concerted evolution, meaning that they show a high level of sequence conservation within a species but are divergent between species. However, while their sequences are divergent, the function of centromeres is conserved throughout eukaryotic biology and possibly based on a universal chromatin structure196. Telomeres are the nucleoprotein complexes that occur at the ends of eukaryotic linear chromosomes. In the mammalian genome they consist of several kilobase pairs of repetitive DNA sequences (TTAGGG) that attract a number of sequence- and structurespecific binding proteins. These chromosomal caps prevent nucleolytic degradation and provide a mechanism for cells to distinguish natural termini from DNA DSBs, which signal DNA damage and would result in cell-cycle arrest, senescence, or apoptosis197. The protective properties of telomeres were recognized by Barbara McClintock and Herman Muller, the latter of whom coined the term telomere. Telomeres terminate in 35–600 bases of single-stranded TTAGGG at the 3' end (the 3' overhang). This 3' overhang folds back into the duplex TTAGGG repeat array, forming a so-called T loop198 (see also Fig. 4.7a). Telomeres are also thought to buffer the internal coding regions of the genome from the consequences of the end replication problem; that is, the inability to complete the 5' end by lagging-strand synthesis (Fig. 4.7b). While this may temporarily protect the genome against attrition, cells would inevitably lose terminal DNA with each cell division. However, telomere attrition can be countered by elongation mechanisms (see also Chapters 4 and 6).
3.1.5 NON-CODING RNA GENES Non-coding RNA genes produce transcripts that function directly in regulatory, catalytic, or structural roles in the cell. They represent a major component of the transcriptomes of
GENOME STRUCTURE AND FUNCTION
69
higher organisms and can be subdivided into housekeeping RNAs and regulatory RNAs199. Housekeeping RNAs include rRNAs, tRNAs, small nuclear RNAs, and small nucleolar RNAs, implicated in such functions as splice regulation and rRNA modification. Also the template RNA component of telomerase (TERC) should be considered as a housekeeping non-coding RNA. Regulatory non-coding RNAs include miRNAs, which can induce posttranscriptional gene-silencing activity, either by translational repression or by triggering mRNA degradation (Chapter 1). In addition to the small non-coding RNAs, there are also larger regulatory non-coding RNAs. The best known example is the X-inactive specific transcript (Xist) gene. The 15–17-kb-long (in mice) Xist non-coding RNAs play a key role in transcriptional silencing of the X chromosome during early female embryogenesis, as part of the dosage compensation of X-linked gene products in mammals200. At the time of X-chromosome inactivation, Xist RNA becomes stable and accumulates. It spreads and eventually coats the whole chromosome. Whereas it is not exactly clear how this spreading takes place and how it brings about transcriptional silencing, evidence has been obtained that L1 elements, which are enriched on the X chromosome, serve to promote spreading. As mentioned above, methylation of the L1 repeats may play a role in stabilizing the inactive state, which originally may have been a silencing mechanism to defend the genome against parasites191. Modification of histone proteins may play a role as well (see below for a discussion of the role of methylation and histone modification in gene silencing).
3.1.6 MITOCHONDRIAL GENOMES In a discussion of genome primary structure focused on mammals and other eukaryotes, it is important to mention the existence of a separate genome: the mitochondrial genome. Neither eubacteria nor archaea contain organelles, such as mitochondria, lysosomes, and endoplasmatic reticulum, which abound in eukaryotes. According to the endosymbiosis theory, the mitochondria of eukaryotes evolved from aerobic bacteria (probably related to the rickettsias) living within their host cell201. Mitochondria can arise only from preexisting mitochondria and cannot be formed in a cell that lacks them because the nuclear genome encodes most but not all of the mitochondrial proteins. The remainder is encoded in the mitochondrion’s own genome, which resembles that of prokaryotes, not that of the nuclear genome. The genome of human mitochondria contains 16 569 bp of DNA organized in a closed circle (Fig. 3.6). This encodes two rRNA molecules, 22 tRNA molecules and 13 polypeptides, which participate in building several protein complexes embedded in the inner mitochondrial membrane: seven subunits that make up the mitochondrial NADH dehydrogenase, three subunits of cytochrome c oxidase, two subunits of ATP synthase, and cytochrome b. Vertebrate transcription is initiated at two promoters, PH and PL for heavy and light strands respectively, located 150 nucleotides apart
70
GENOME STRUCTURE AND FUNCTION PH
PL D-loop
rRNA 12S
Cyt b
l Va
t RN A Gl tRNA Pro tR u NA Th r
tRN y Gl
ND
tRNA
Lys
3
n As
A
s Cy
A tRN
Ala
tRNA Asp
tR
fMet
A tRN
NA
tRNA
Gln
tRNA Try t RN A
g
tR
ND4
Trp
tRNA His
Ser(AGy)
NA
CUN) Leu(
A
tR tRN A Ar
N
2 ND
tRNA
tRNA Leu (UU R) tRNA IIe
NA tR
tRNA
ND1
A
5 ND
16 Sr RN
ND 6 Phe
CN )
tRNA Ser (U
CO I
lll CO A6
COII
Fig. 3.6 Organization of the 16 569-bp human mitochondrial genome.
within the D-loop regulatory region. The H- and L-strand transcriptional units specify multiple genes, as in prokaryotes. However, like in eukaryotes, the mRNAs have poly(A) tails. The proteins encoded by nuclear genes (e.g. cytochrome c and the RNA and DNA polymerases used within the mitochondrion) are synthesized in the cytosol and then imported into the mitochondrion.
3.1.7 SUMMARY In summary, the primary chemistry of the DNA of the genome dictates a highly stable structure harboring an abundance of functional elements of which protein-coding sequences are only a very small fraction. Most of the functional elements in a genome are probably involved in transcription regulation or in facilitating chromosome structure. How do such sequence elements exert their function? It is now clear that many of the functional DNA elements in the genome function through alterations in DNA higherorder structure, often through interactions with proteins. This is obvious for those sequences determining specific chromosomal structures relevant for processes such as
GENOME STRUCTURE AND FUNCTION
71
cell division. However, the packaging of the DNA in the cell also controls accessibility of the sequences essential for basal or regulated gene expression. In Chapter 4 it will become clear that alterations in DNA packaging are important, not only for transcription regulation and basic structural maintenance, but also for the continuous repair of chemical damage inflicted upon the DNA from a multitude of endogenous and environmental sources. While attention thus far has been focused on alterations in DNA primary structure and sequence, changes in DNA higher-order structure, also termed epigenomic changes, are gaining increasing interest as possible causal factors in aging and disease (see also Chapter 6). Below, I will briefly discuss DNA higher-order structure and the mechanisms that facilitate genome reprogramming (or epigenetic control) during development, in tissue-specific gene expression, and in global gene silencing.
3.2 Higher-order DNA structure One level up in the structural regulation of genome function from the DNA double helix and its primary sequence code is the folding of its DNA into chromatin. Chromatin is more than the genome and denotes the complex of DNA and proteins that form the chromosomes. As we shall see, by organizing itself as chromatin the genome maximizes its information content and simultaneously creates ways to make this information available whenever and wherever this is needed. Most prokaryotic genomes are contained in a single, circular DNA molecule, tightly coiled in a compact structure called the nucleoid, with much less associated protein than eukaryotic chromosomes. The genome of eukaryotes is organized in the form of linear chromosomes, composed of euchromatin and heterochromatin, residing in a nucleus separate from the cytoplasm. Heterochromatin is compact, generally inaccessible to DNA-binding factors and transcriptionally silent. Euchromatic domains are the more accessible and transcriptionally active portions of the genome. What is the underlying structure that defines chromatin at the functional level? The basic unit of chromatin is the nucleosome. Nucleosomes consist of approximately 200 bp of DNA wrapped around an octamer of core histone proteins, consisting of two copies each of histones H2A, H2B, H3, and H4. This structure is the beads-on-a-string-conformation and can be condensed into the tightly packaged solenoid structure or 30-nm fiber. The molecular arrangement of the higher levels of chromatin organization is not well understood, but light- and electron-microscopic studies of interphase and metaphase chromosomes have revealed fibers ranging from 100 nm to the 700 nm structure seen in the metaphase chromosomes during cell division202,203. Figure 3.7 schematically depicts what is probably the current consensus with respect to the organization of the DNA in a mammalian cell nucleus.
72
GENOME STRUCTURE AND FUNCTION DNA
Histone core
30 nm fiber
Metaphase chromosome
Interphase nucleus
Fig. 3.7 Higher-order structure of nuclear DNA into chromatin.
Active or potentially active genes exist in the simple nucleosomal structure (euchromatin). In the regulatory regions of active genes nucleosomes are either removed or undergo structural alteration, which facilitates the binding of transcription factors (see below). These segments are sensitive to chemical and enzymatic attack because they are protected by histone proteins only poorly, if at all. They are called hypersensitive sites to indicate their extreme sensitivity to enzymatic digestion and typically appear and disappear in patterns that are coordinated with gene activity; that is, more hypersensitive sites appear as gene activity increases. The highly compacted chromatin fibers represent heterochromatin domains, where genes are not transcriptionally active. Some heterochromatin is condensed chromatin that unfolds and becomes transcriptionally active during some portion of the cell cycle. Constitutive heterochromatin remains transcriptionally inert during the entire cell cycle. The bulk of constitutive heterochromatin is composed of repetitive DNA, such as at
GENOME STRUCTURE AND FUNCTION
73
the centromeres and telomeres. Because DNA replication requires polymerases and regulatory proteins similar to those required for DNA transcription, condensed heterochromatin most likely unfolds into euchromatin before replication proceeds. Using powerful new methods the genomic distribution of nucleosomes is now being determined. In the ChIP-on-chip method204, protein–DNA complexes are cross-linked by the addition of formaldehyde to living cells. After lysis of the cells and mechanical shearing of the chromatin to yield fragments of 0.5–2 kb, the cross-linked protein–DNA complexes are immunoprecipitated using antibodies against invariant portions of histones. The DNA, recovered from the immunoprecipitated complexes, is then used as a probe to hybridize a DNA microarray harboring thousands of sequences covering the genome. The microarray signals indicate nucleosome density at that particular sequence. Application of such genome-wide methods is easiest in organisms with a small genome, such as yeast. Indeed, for this organism it could be concluded that upstream regions of highly active genes in yeast display a reduced nucleosome density compared to the upstream regions of inactive genes205. Hence, gene activation in yeast is associated with reduced nucleosome density. How is the situation in the genome of higher eukaryotes, which is so much more complex? As mentioned above, human genes are not uniformly spread across the genome, but tend to be clustered together. To determine the chromatin architecture of the human genome, Gilbert et al. separated compact- and open-chromatin-fiber structures from a human lymphoblastoid cell line by sucrose sedimentation and analyzed their distributions by hybridization to metaphase chromosomes and genomic microarrays206. Their results indicate that gene-rich areas in the human genome are preferentially located in open chromatin fibers; that is, structures mostly devoid of fibers beyond 30 nm. However, this first study to globally map higher-order DNA structure in the human genome also revealed that there is no strict correlation between open chromatin and the activity of a gene. Indeed, not every gene in the open areas is being transcribed and active genes in regions of low gene density can be embedded in compact chromatin fibers. As will be outlined later, various protein factors are able to regulate transcriptional activity through chromatin remodeling at the level of an individual gene. Various factors are involved in maintaining and altering higher-order DNA structure, which is exquisitely regulated and provides an additional layer of information to the primary sequence code of the DNA. This type of regulation involves epigenetic alterations, heritable changes in gene function that occur without a change in the DNA sequence207. The best-known example is genomic imprinting, which is a type of genomic modification that in mammals dictates the inactivation of either the paternal or maternal copies of a gene. Methylation is the epigenetic marker most likely responsible for repressing the activity of one of the alleles. Methylation of C residues at carbon 5 of the pyrimidine ring, primarily at CpG sequences, is widespread in mammalian genomic DNA and essentially an extension of the genetic code to a fifth base. Imprinted genes often contain
74
GENOME STRUCTURE AND FUNCTION
parent-specific differences in DNA methylation within genomic regions known as differentially methylated domains (DMDs). Patterns of CpG methylation are maintained by the action of Dnmt1, the mammalian maintenance cytosine methyltransferase enzyme. Apart from imprinted genes, other regions of the genome with high levels of methylated CpG dinucleotides are the inactive X chromosome in female mammals and mobile genetic elements, all of which are associated with stable transcriptional repression. Gene silencing often occurs through hypermethylation of CpG-rich, promoterassociated regions, termed CpG islands (see below). Yet another genome project, the Human Epigenome Project (HEP), aims to identify, catalogue and interpret genome-wide DNA methylation patterns of all human genes in all major tissues. The data, obtained by high-throughput methods, are deposited in a public database (www.epigenome.org). Since methylation patterns are known to undergo alterations during aging, such a systematic whole-genome study of DNA methylation at the sequence level could be a prelude to the systematic evaluation of agingrelated patterns of DNA methylation levels of sites within the vicinity of the promoter and other relevant regions of a gene (see also Chapter 6). Other factors that determine epigenetic regulation of gene expression involve histone proteins. DNA is bound to the histones through electrostatic forces between the negatively charged phosphate groups in the DNA backbone and positively charged amino acids (e.g. lysine and arginine) in the histone proteins. Histone proteins can be modified by the addition of acetyl, methyl, or phosphate groups, and this alters the strength of the bonding between the histones and DNA. Modifications such as these are usually associated with the regulation of DNA transactions such as replication, gene expression, chromatin assembly and condensation, and cell division. It has been suggested that, together, these modifications may form a complex, regulatory code: the so-called histone code. Attempts to decipher this code are currently underway, using the aforementioned ChIPon-chip method with antibodies against specific histone modifications. The first results of this global analysis of histone modifications, in yeast and Drosophila, suggest very similar genomic distributions of virtually all tested histone modifications; for example, H3 and H4 acetylation and H3K4 di- and tri-methylation208. Moreover, most modifications appeared to be positively correlated with gene expression. This suggests that the different histone maps are linked, pointing towards the absence of a complex regulatory code of histone modifications. Multiple histone modifications employed in parallel to control transcription might be another example of redundancy, conferring robustness on transcriptional regulation. On the other hand, it is too early to rule out a role of histone modification patterns in the timing of specific gene activities. How does the cell remember its epigenetic marks during cell division and how does it remodel the epigenetic modifications inherited from the transcriptionally inactive sperm and egg? Before a cell can divide, it must duplicate its DNA. In eukaryotes, this occurs during the S phase of the cell cycle. The process is schematically depicted in
GENOME STRUCTURE AND FUNCTION
75
Origin of replication
5'
3' Leading strand Lagging strand Okazaki fragment
3'
5'
Replication fork movement
Fig. 3.8 DNA replication is bidirectional and semi-discontinuous, with the leading strand synthesized in the 5’ → 3’ direction as a single, continuous strand and the lagging strand synthesized discontinuously, but also in the 5’ → 3’ direction.
Fig. 3.8. After a portion of the DNA is unwound by a helicase, a molecule of DNA polymerase binds to one strand of the DNA, moving along in the 3' → 5' direction to assemble a leading strand of nucleotides and reforming a double helix. Then, a second DNA polymerase (polymerase in eukaryotes) binds to the other template strand as the double helix opens. It has to work the other way, because DNA synthesis can only occur 5' → 3'. This molecule must synthesize discontinuous segments of polynucleotides (called Okazaki fragments). Another enzyme, DNA ligase I, then joins these together into the lagging strand. Both leading and lagging strands need primers, generated by a primase. The two DNA polymerases carrying out leading- and lagging-strand synthesis are locked together in a replication machine with the DNA template for the lagging strand looping out from the twin polymerases. This mode of replication, predicted by Watson and Crick and so elegantly confirmed experimentally by Meselson and Stahl (Chapter 1), is described as semi-conservative: one-half of each new molecule of DNA is old and one-half is new. Note that synthesis using the 5' → 3' strand as the template presents a special problem, which reveals yet another vulnerability of the genome for aging-related alterations. As the replication fork nears the end of the DNA, there is no longer enough template to continue forming Okazaki fragments. So the 5' end of each newly synthesized strand cannot be completed and each of the daughter chromosomes will have a shortened telomere. It is estimated that each mitosis event costs human telomeres about 100 bp of DNA. This so-called endreplication problem was recognized by Alexey Olovnikov in 1971, and independently by James Watson a year later209. An average 150-million-bp human chromosome can be copied within an hour because of the many places where it can begin. Indeed, whereas bacteria have one single
76
GENOME STRUCTURE AND FUNCTION
specific origin of replication, eukaryotes can start replication at multiple origins on each chromosome. DNA replication proceeds bidirectionally from an origin of replication; that is, in opposite directions away from the origin (Fig. 3.7). At each cell cycle, during S phase, duplication of chromatin structure, which involves both redistribution of parental histones and histone neosynthesis, occurs in tight coordination with DNA replication210. There are several ways to maintain DNA chromatin organization. First, during DNA replication there are differences in replication timing. In general heterochromatin replicates later than euchromatin. It is conceivable that partitioning the genome into domains with specific replication times helps the cell to remember how to distribute histones between the daughter strands. Second, specific proteins, such as the chromatin assembly factor-1 (CAF-1) complex, help in the assembly of nucleosomes onto newly replicated DNA. In this process, the proliferating cell nuclear antigen (PCNA) plays an important role as a sliding clamp, serving as a loading platform for many proteins involved in DNA replication (and DNA repair; see Chapter 4). Once DNA and chromatin replication is complete, nuclear division follows (see below). Methylation patterns are perpetuated after replication by a maintenance methylase, DNA methyltransferase 1 (DNMT1211). Its importance is illustrated by the observation that inactivation of this gene in the mouse is lethal. The enzyme acts on the hemimethylated sites and converts them into fully methylated sites. Once DNA and chromatin replication are complete, nuclear division follows (see below). In the cell divisions that give rise to the zygote, a crucial step is the decondensation and reorganization of chromatin of male and female gametes. This basically involves a largely genome-wide erasure of the germ-line-specific epigenetic modifications that had occurred during normal development in the embryonic primordial germ cell lineage212. The identification of epigenetic reprogramming mechanisms is a major current interest in view of the fact that successful cloning of animals from differentiated adult cells by somatic cell nuclear transfer also requires nuclear reprogramming. The latter is less successful than the natural reprogramming of gamete DNA, as testified by the widespread epigenetic defects in nuclear-transfer embryos213. A detailed discussion of this, as-yet insufficiently explored, topic is beyond the scope of this book. Of note, while occurring with exceptionally high fidelity, the entire process of information transmission during cell division, including gametogenesis and embryonic development, is unlikely to be faultless, even under normal conditions in the absence of DNA chemical insults or toxicants. Errors are minimized by different and overlapping mechanisms (see Chapter 4), but cannot be entirely prevented. As mentioned in Chapter 2, in germ cells, the DNA mutations resulting from such errors are the driving force behind evolutionary change. Indeed, in times of stress, the propensity to mutate and to rapidly create variants that can escape selection pressures facilitates survival of a small fraction of the original population. However, in somatic cells they may lead to cellular degeneration and death (Chapter 6).
GENOME STRUCTURE AND FUNCTION
77
3.3 Nuclear architecture After this brief discussion of the primary and higher-order structure of the DNA of the genome, it is important to review the physical space within which this complex set of structure–function relationships unfolds. Evidence is rapidly emerging for a critical role of nuclear architecture in determining gene activity. Nuclear architecture involves the chromosomal positioning in the nuclear space. We have already seen that in interphase cells the DNA of the genome is organized as linear chromosomes, which are really looped 30-nm solenoid fibers, alternating with beads-on-a-string structures. How is this organized in the nuclear three-dimensional space and what are the functional consequences of this organization? The nucleus in eukaryotic cells is separated from the cytoplasm by a nuclear envelope (nuclear membrane). The nuclear envelope consists of inner and outer membranes separated by a perinuclear space (Fig. 3.9). The outer nuclear membrane is continuous with the endoplasmic reticulum and has ribosomes attached. So the space between the inner and outer membrane is directly connected to the lumen of the endoplasmic reticulum. In this way, ribosomal subunits and mRNA transcribed from genes in the DNA can leave the nucleus, enter the endoplasmic reticulum, and participate in protein synthesis. Nuclear pores are formed at sites where the inner and outer membranes of the nuclear envelope are joined. They allow free diffusion of small molecules as well as active transport through
Chromatin
Nucleoplasm
Ribosome Nucleolus Endoplasmic recticulum
Nuclear pore
Lamina network Outer nuclear membrane
Inner nuclear membrane
Fig. 3.9 Schematic depiction of the nucleus of a mammalian cell. Note that the lamins are not only restricted to the inner nuclear membrane but also form a network inside the nucleus, which may be involved in DNA transactions, such as replication and transcription. The dark gray areas represent the chromatin.
78
GENOME STRUCTURE AND FUNCTION
the interaction between import or export signals (nuclear localization signals, or NLSs, and nuclear export signals, or NESs, respectively) with receptors termed importins and exportins. The fluid within the nucleus is called nucleoplasm, which contains the linear chromosomes. The inner nuclear membrane is linked—by integral proteins—to the lamina, a fibrous meshwork composed of intermediate filament proteins called lamins (lamins A/C and B). Lamins A and C arise from the same gene, LMNA, by alternative splicing. The lamins are coiled-coil structures that contain a small N-terminal head followed by a rod-like domain (␣-helical coiled-coil) and a C-terminal globular tail.Via the coiled-coil regions, lamins can form parallel dimers, which in turn form polymers with other lamin dimers in an antiparallel manner (head-to-tail). The nuclear lamina is dynamic and can depolymerize during mitosis and reform upon re-entry into interphase following rounds of phosphorylation and dephosphorylation at residues flanking the lamin coiled-coil domains214. Lamins at the nuclear periphery are thought to maintain nuclear shape. This becomes apparent after treatment of cells with non-ionic detergents, a treatment that solubilizes the vast majority of cytoplasmic and nuclear proteins. In such cases, electron-microscopic images indicate that the nucleus of the treated cells retains its shape and the lamina remains at the nuclear periphery215. Lamins are also dispersed throughout the nucleoplasm, forming a thin fibrillar network proposed to be major structural elements of the internal nuclear matrix216 (Fig. 3.9). Heritable mutations in lamin A/C or lamin-binding proteins cause various diseases. For example, mutations in emerin (a lamin-binding protein) or A-type lamin cause Emery–Dreifuss muscular dystrophy. Mutations in LMNA can also cause Dunnigan-type partial lipodystrophy or Hutchinson–Gilford progeria syndrome (HGPS; see Chapter 5). The mutations that cause these diseases are non-overlapping and it is still unclear as to how different mutations in this gene can cause different diseases217. However, this is not unprecedented. For example, mutations in the RET proto-oncogene can cause multiple endocrine neoplasia and Hirschsprung disease218 and we will see in Chapter 4 that different mutations in the XPD helicase gene can cause different diseases as well. Lamins may contribute to tissue-specific gene expression, through their role as key elements in nuclear architecture. Regions of chromatin appear to be anchored to the lamina and, at least in vitro, lamins have been demonstrated to bind directly to chromatin. As discussed in more detail in Chapter 5, children with HGPS are born normal but start developing multiple symptoms of premature aging within 1 year. The gene product resulting from the dominant mutation that gives rise to this disorder disrupts nuclear architecture, which in a way that is still obscure is ultimately the cause of the disease. Interestingly, in both human skin fibroblast cell lines from old individuals219 and nonneuronal tissues from aged nematodes220 progressive deterioration of nuclear lamina and chromatin architecture has been observed. In the nematodes this degeneration was delayed in the long-lived daf-2 mutants and accelerated in the short-lived daf-16 mutants.
GENOME STRUCTURE AND FUNCTION
79
Whereas there is no complete understanding as to how transcription or other DNA-dependent transactions, such as replication and repair, are distributed within the nucleus, evidence is emerging that none of this is random. For example, whereas originally it was thought that the replication machinery was moving along the stationary DNA, it is now considered more likely that replication is organized at fixed positions in so-called replication factories221. Likewise, various models have been proposed in which RNA is transcribed on a solid substrate with an aggregate of all enzymes needed for transcription, processing, and transport in place. Such a transcription-factory model implies that different genes do not always assemble their own transcription sites de novo when they become active, but instead migrate to such a pre-assembled association. Compartmentalization of DNA-dependent transactions in the three-dimensional space of the nucleus would be in keeping with the organization of other cellular processes, such as the electron-transport chain in mitochondria, and is certainly a much more likely model than the assumption that all these processes happen at random in solution. A good candidate for such a solid support is the nuclear matrix, also termed nuclear scaffold, nuclear cage, or nuclear skeleton. The nuclear matrix is a meshwork of proteins that connects to the cytoskeleton at the nuclear envelope. We have already seen that lamins could play a structural role in this meshwork. It has been suggested that in the interphase nucleus the DNA, organized in loops, is anchored to the nuclear matrix by means of non-coding sequences known as matrix-attachment regions. These matrix-attachment regions may constitute boundaries of independently controlled chromatin units within which the DNA packaging and function may be changed without affecting the neighboring regions (see also below). However, the existence of a nuclear matrix is controversial and an extensive meshwork of filaments in the interchromatin space could simply be an experimental artifact222. In contrast to the biochemically well characterized lamina, little is known about the protein composition of the internal nuclear matrix. Others, however, argue that a nuclear matrix is a readily observed cellular structure and that the concept of a matrix to provide architectural support for higher-order chromatin packaging is sound223. It is conceivable that chromatin is itself responsible for nuclear structure. In this respect, emerging evidence suggests that the distribution of chromosomes in mammalian interphase cell nuclei is non-random. Whereas the physically distinct nature of chromosomes is clearly visible during mitosis—when chromosomes condense and appear as separate entities—the technique of chromosome painting demonstrated that also during interphase each chromosome occupies a well-defined nuclear sub-volume, called chromosome territory224. In chromosome painting, probe sets for specific chromosomes, labeled with different fluorescent dyes, are hybridized to metaphase plates or interphase cells so that each of the chromosomes shows a different color when viewed with a fluorescence microscope. Chromosome territories are non-randomly arranged within the nuclear space and occupy preferential positions relative to the center of the nucleus
80
GENOME STRUCTURE AND FUNCTION
and relative to each other.Work by Tom Misteli and co-workers, who carried out a systematic analysis of the spatial positioning of a subset of mouse chromosomes in several tissues, suggests a pattern that is tissue-specific225. The genes within chromosome territories appear to be non-randomly positioned relative to each other, or to nuclear landmarks, such as the nuclear envelope. An attractive model has been presented in which the clustering of genes or other functional sequences into contiguous regions of the three-dimensional space of the nucleus, termed neighborhoods, is explained by similar requirements for optimal function, such as coordinated and efficient expression226. As yet, it is unclear whether this model is generally applicable, but there are some intriguing examples of such potential compartmentalization. For example, ribosomal genes cluster together in physical space due to the congregation of the various chromosomes harboring the tandem arrays in which these genes occur, to form the so-called nucleoli (Fig. 3.8). Each nucleolus contains genetic material from multiple chromosomes. An area of DNA called the nucleolar organizer directs the synthesis of rRNA, which subsequently combines with ribosomal proteins to form immature ribosomal subunits that mature in the cytoplasm after they leave the nucleus by way of the pores in the nuclear envelope. Apart from nucleoli, other such compartments include Cajal bodies, a congregation of small nuclear RNA and histone gene clusters, nuclear speckles or spliceosomes, the so-called promyelocytic leukemia (PML) nuclear bodies and, possibly, repairosomes. Interestingly, the gene-neighborhood model would indicate an additional concept of gene position effects. That is, the actions of a gene are not only influenced by its position in the one-dimensional DNA sequence, relative to regulatory elements, but also by its particular location in the nucleus. Position effects here may include not only transcription, but also replication, repair, and recombination. For example, when DNA undergoes more than one DSB the spatial proximity of the broken ends is positively correlated with the probability of illegitimate joining227. Similar to the transmission of the DNA sequence and its methylation patterns and histone code, nuclear compartmentalization must also be accurately perpetuated during cell division to prevent cell death or phenotypic change.After generating two identical chromosomes, termed sister chromatids, which remain attached at the centromere, nuclear division begins with the prophase. In this stage the nucleolus disappears and the nuclear envelope fragments. The latter happens, as mentioned already, by disassembling nuclear lamins, which is regulated by their phosphorylation. The mitotic spindle begins to assemble as the two centrosomes (duplicated just before mitosis) migrate away from each other until they are on opposite sides of the nucleus. Each centrosome contains a pair of centrioles, consisting of nine microtubule triplets surrounding a hollow cylinder. The centrioles are replicated at the same time as the DNA is replicated, also in a semi-conservative manner; that is, a daughter centriole grows out of the side of each parent centriole. The centrioles may function to orientate the spindle and anchor the microtubules radiating out to the
GENOME STRUCTURE AND FUNCTION
81
chromosomes with some of them extending to the other pole. The point of connection on each chromosome is the kinetochore, a protein complex in the constricted regions defined by the centromere. The centromeric satellite repeat sequences discussed above bind to the kinetochore through specific proteins228. During pro-metaphase the chromosomes are attached to the spindle and move to align at the metaphase plate or equator of the spindle. During metaphase, nucleus and chromatin have essentially become spindle and sister-chromatid pairs, with the DNA at its highest level of compaction. Mitosis has now reached the essential point, to correctly distribute identical chromosomes over the future daughter cells by pulling the sister chromatids of each pair toward opposite poles. This process is guided by a quality-control mechanism called the mitotic spindle checkpoint. This checkpoint allows every chromosome to send a stop signal, arresting cell growth until all the chromosomes are appropriately distributed. Defects in this checkpoint provoke chromosome mis-segregation and aneuploidy (gain or loss of chromosomes), which can have adverse functional consequences that include cell death and cancer. For example, defects in different components of the mitotic spindle checkpoint have consistently been observed in cancer cells, characterized by chromosomal instability. As described in detail in Chapter 5, partial inactivation of genes encoding proteins of the checkpoint machinery cause an increase in aneuploidy and are associated with increased cancer and symptoms of accelerated aging. In the anaphase, then, the two sister chromatids separate at the centromere and move to opposite poles, driven by the depolymerization of microtubules at the kinetochores. Finally, in the telophase the spindle disappears, the chromosomes decondense and the nucleoli reappear. Also the nuclear envelope reforms and at the same time the lamins are dephosphorylated and reassembled. It is unclear as yet what the exact role of lamins is in assembly of the nuclear envelope.
3.4 Transcription regulation The elaborate three-dimensional structure of our genome equipped with sophisticated tools for the transfer of the information it contains, as described here, has developed over the last 3–4 billion years since the emergence of our ancestors, the first replicators. Its purpose has never been any other than to make its information available for providing the functions that ultimately serve its own perpetuation. This has resulted—through the inevitable errors in the transmission of genetic information—in the spectacular level of evolutionary diversity that can be witnessed all around us. The information encoded in the genome is not solely the digital information specifying the amino acid sequences that underlie the complicated protein machines that run our cells. More importantly, the code dictates the time and place of protein expression as well as the interactions that result in
82
GENOME STRUCTURE AND FUNCTION
the regulatory networks that control the function of the molecular machines in the first place. Whereas protein expression can be regulated at various levels and through different means, control at the gene-transcriptional level is by far the most important. It is at this level of regulation where decisions are made as to why some proteins are expressed in one tissue and not in others, why particular stimuli activate expression of a certain protein and have no effect on others, and how the complex protein machines come together, often acting in concert with many others, to execute their functions. It is also at this level where we can expect to find the most significant adverse effects of the aging process, for example, as a consequence of genomic alterations due to DNA damage or errors during information transfer. Our knowledge of transcription in the relatively simple prokaryotes is much deeper than of that in eukaryotes, which have to operate in a less gene-dense genome. Nevertheless, it is clear that in both types of species transcription regulation is fundamentally similar, with more complexity and intricacy in eukaryotes. In prokaryotes transcription and its regulation is dominated by operons, groups of adjacent, co-expressed, and co-regulated genes that encode functionally interacting proteins (discovered by François Jacob and Jacques Monod in the early 1960s). Operon transcripts always code for more than one protein, and prokaryotes can start translation of the mRNA into amino acids separately at the beginning of each protein-coding section. By contrast, in eukaryotes, mRNA is not directly used in protein synthesis, but must undergo processing, including the removal of introns, and the addition of a 7-methylguanylate cap structure at the 5' end and a poly(A) tail at its 3' end, prior to export from the nucleus to the cytoplasm, where it is attached to the ribosomes for translation. Whereas functionally related genes in eukaryotes are often clustered, eukaryotic transcription machinery generally cannot handle polycistronic transcripts. Operons are rare in eukaryotes, but the nematode C. elegans occasionally does make polycistronic transcripts and can process them229. The myriad of processes that regulate transcription, in both prokaryotes and eukaryotes, all converge on the promoter, a term introduced in 1964 by Jacob, Ullman, and Monod230 for a site on DNA that is upstream (5') to coding sequences to which RNA polymerase will bind and initiate transcription. In contrast to prokaryotes, which require only one kind of RNA polymerase, eukaryotes require three RNA polymerases: RNA polymerase I synthesizes rRNA (90% of the RNA in a cell); (2) RNA polymerase II synthesizes pre-messenger RNAs; and (3) RNA polymerase III is responsible for the synthesis of tRNA, 5 S RNA, and small nuclear RNA. RNA polymerase II is the most delicately regulated eukaryotic RNA polymerase with an extensive need for accessory proteins. In prokaryotes the promoters of genes are approximately 200 bp long and consist of two conserved regions, called 35 and 10, because they are approximately 35 and 10 nucleotides upstream from the transcription initiation site, which is called 1 (base number 1). The 10 region contains the so-called Pribnow box with a sequence similar
GENOME STRUCTURE AND FUNCTION
83
GTFs/RNA Pol II
Pre initiation complex Gene TATA
INR
DPE
30
1
30
Fig. 3.10 Schematic representation of a core promoter of a mammalian gene. A preinitiation complex is assembled at the core promoter through the binding and interaction of the general transcription factors (GTFs) and RNA polymerase II (RNA Pol II). DPE, downstream promoter element; INR, initiator element; TATA, TATA box. See text for details.
to 5'-TATAAT-3', which is one of the signals to initiate transcription. The expression level of metazoan genes is regulated by the core promoter, cis-acting sequences, and trans-acting factors. A prototypical metazoan promoter is assembled from a modular array of relatively short sequence motifs (about 7–20 bp), each of which independently represents a binding site for specific transcription regulatory proteins. The core promoter contains DNA sequence elements that are recognized by the general transcription machinery, which help to direct and orient the preinitiation complex at the promoter and play a critical role in the regulation of transcription (Fig. 3.10). The best characterized element is the TATA element, an AT-rich sequence located about 25–35 bp upstream of the transcription-initiation site (equivalent to the Pribnow box in bacteria), which is recognized by the TATA-binding protein subunit of the transciption factor IID TFIID, initiating preinitiation-complex formation. Other elements include the pyrimidine-rich initiator element (INR), spanning the transcription start site, and the downstream promoter element (DPE), which is recognized by components of TFIID other than TATA-binding protein. Not all promoters contain all these elements. For example, the DPE is often present in promoters that do not contain a TATA element. The strongest core promoters contain both TATA and INR elements. Weaker core promoters, such as promoters of housekeeping genes, expressed in all tissues at a low level, generally lack a TATA element, an INR element, or both. Such differences in core promoter do not affect the assembly of the RNA polymerase II preinitiation complex. Bacterial RNA polymerase can initiate transcription in vitro from a core promoter. This is in contrast to eukaryotic RNA polymerases, which need initiation factors, termed general transcription factors, that are required to assemble the stable transcriptional complex needed for all three eukaryotic RNA polymerases. Of these three polymerases, RNA polymerase II is associated with a wide variety of regulatory events affecting the genes transcribed by this enzyme. Also the general transcriptional complex for RNA polymerase II contains far more components than the two other RNA polymerases. The first step in the formation of this complex is the binding of TFIID to the TATA box or
84
GENOME STRUCTURE AND FUNCTION
equivalent region, which is facilitated by TFIIA. TFIIB then joins the complex by binding to TFIID, allowing the recruitment of RNA polymerase to the complex, in association with TFIIF. Following polymerase binding, two other general transcription factors, TFIIE and TFIIH, associate with the complex. TFIIH, which has a helicase activity, then unwinds the double-stranded DNA, thereby permitting its being copied into RNA. TFIIH is also important in DNA repair and genetic defects in the helicase component of this and other proteins in mammals have been demonstrated to cause defects in DNA-repair processes and are often associated with the premature appearance of symptoms of aging (see Chapter 5). In metazoa, recruitment of the general transcription factors to the core promoter completes the formation of the preinitiation complex. This allows low levels of accurate transcription in vitro, but not in vivo. In vivo transcription utilizes binding sites for transcription factors231, which can function as activators or repressors. Whereas the core promoter with its preinitiation complex of RNA polymerase II provides the basic machinery of transcription, it is the ensemble of transcription factors that orchestrates the patterns of gene expression in an organism. Within eukaryotes there is a great variety in the number of transcription factors, reflecting the differences in complexity. In yeast, there are about 300 transcription factors, including subunits of general transcription complexes such as TFIID. C. elegans and Drosophila have at least 1000 transcription factors, with as many as 3000 in humans. Transcription factors are structurally organized as combinations of a regulatory domain (for activation or repression) with a DNA-binding domain. Examples of DNA-binding domains are the helix-turn-helix domain, the homeodomain, the zinc-finger domain, the winged helix domain, and the leucine zipper domain. Many transcription-factor-binding sites are within about 200 bp upstream of the core promoter, an area referred to as the proximal promoter. There are binding sites here for at least three types of transcription activation. First, constitutive elements, such as Sp1, bind transcription factors for ubiquitously expressed genes, such as housekeeping genes. Second, inducible elements bind transcription factors that are involved in responses to intra- or extracellular signals, such as the heat-shock-response element and steroidresponse elements. Third, tissue-specific elements bind transcription factors that regulate the expression of tissue-specific genes. An example is the CArG element, a motif repeated four times within the proximal promoter of the human cardiac actin gene and responsible for heart- and muscle-specific transcription. Other cis-acting sequences are often much further away from the core promoter. Examples are enhancers, which are sequence elements located upstream, downstream, or within a transcription unit that can influence the level of gene expression by increasing the activity of a promoter. A special group of enhancer elements is a so-called locuscontrol region, which regulates the expression of functional gene clusters, such as the
GENOME STRUCTURE AND FUNCTION
85
Co-activators Transcription factors
Enhancer Boundary element
Co-repressors
Proximal Core promoter promoter
Exon Intron Enhancer
Boundary element
GTFs/RNA Pol II
Fig. 3.11 Schematic representation of the transcription unit of a typical mammalian gene. Indicated are the introns and exons of the gene, its core promoter and various cis-elements that contribute to the regulation of gene expression. GTF, general transcription factor; RNA Pol II, RNA polymerase II.
-globin gene cluster in the red-blood-cell lineage. Finally, there are silencer elements, which act to inhibit gene transcription, and insulator or boundary sequences, which prevent enhancers associated with one gene from inappropriately regulating neighboring genes. All these cis-regulatory sequences can be scattered over distances as great as 100 kb in mammals (Fig. 3.11). How do transcription factors activate or repress transcription from the core promoter in the context of all these regulatory sequences? This comes down to the question of how transcription-factor binding to a recognition site that is often some distance away from the core promoter conveys its activation or repression signal to the general-transcriptionfactor-based transcription-initiation complex at the site of the core promoter. Whereas there are some examples of direct interaction between transcription factors and general transcription factors, indirect interaction through co-activators and co-repressors seems to be the norm. Co-activators or co-repressors provide a connection between the transcription factors and the preinitiation complex. There are different classes of transcriptional co-regulator, some intrinsic to components of the core machinery, others facilitating enhancer–promoter contact, sometimes over large distances. This requires looping between a gene and a distal enhancer, which is determined by higher-order DNA structure. This indicates the importance of cofactors that can alter this higher-order structure. Many transcriptional co-activators and co-repressors have now been identified as factors that either covalently modify the N-termini of core histones through phosphorylation, acetylation, or methylation, or actively effect nucleosome positioning. For example, gene-specific transcriptional activation often involves the targeting of two types of chromatin-remodeling enzyme to the core promoter: histone acetyl transferase (HAT) and the ATP-dependent SWI/SNF-like complex232. HATs have been demonstrated to
86
GENOME STRUCTURE AND FUNCTION
mediate transcription activation through acetylation of histones, whereas histone deacetylases (HDACs) can do the opposite. HATs and HDACs may work through the recruitment of other, non-histone chromatin proteins, which subsequently remodel chromatin. SWI/SNF is itself an ATP-driven chromatin remodeling complex that can act to repress a gene or relieve nucleosome-mediated repression of promoters, permitting the assembly of the preinitiation complex. These two types of transcriptional cofactor may act synergistically to establish a local chromatin structure that is permissive for transcription. Apart from transcription factors and their cofactors, two other layers of transcription control are likely to be important. First, DNA methylation at CpG sites has already been mentioned as a major marker for stable transcriptional repression. How does DNA methylation silence genes and how flexible is it? Recent results suggest that the repressive action of methylation may also be ascribed to chromatin remodeling. Likely mediators of methylation-dependent shutdown of gene expression are the proteins belonging to the methyl-CpG-binding domain (MBD) family. These proteins interact specifically with methylated CpG dinucleotides and one of them, MeCP2, has been associated with chromatin-modification enzymes, including HDAC, as well as with the SWI/SNF complex. Binding of MeCP2 near the promoter would lock a gene in a repressive state through chromatin remodeling. Activation of the gene could occur through the loss of the CpG methylation that holds the MeCP2 at the promoter or, alternatively, through the modification of this protein, thereby facilitating its release207. Finally, during the last few years, regulatory RNAs have been linked to transcriptional silencing. The role of the Xist long non-coding RNA in silencing the X chromosome through the promotion of large heterochromatic domains and DNA methylation has already been discussed. Among the small, antisense non-coding RNAs, miRNAs are now rapidly emerging as an extensive regulatory network controlling tissue-specific and development-related gene expression in higher organisms. miRNAs are thought to interact with a mRNA, by binding to partially complementary sites in its 3' untranslated region, and either trigger its destruction or repress its translation. It is possible that as many as 30% of all human genes are regulated by miRNAs233. Some genes can be regulated by multiple miRNAs. However, preventing mRNA translation through direct interaction may not be the only mode of action of miRNAs in transcription silencing. At least in plants evidence has been obtained that miRNAs can act by mediating DNA methylation and heterochromatin formation. The mechanism of this interesting RNA feedback to the genome is unknown, but it may work through the activation of DNA methyltransferase and the recruitment of histone-modifying enzymes. The above makes it clear that transcription is regulated by transcription factors interacting with the core promoter of a gene through the action of a large variety of cofactors. Key mechanisms of transcription regulation are histone modifications and chromatin remodeling, indicating the importance of DNA higher-order structure and chromosomal
GENOME STRUCTURE AND FUNCTION
87
Transcription factory
Genes looping out
Chromosome territories
Fig. 3.12 Chromosome territories with genes looping out to a common transcription factory.
positioning in determining the expression level of a gene. An important aspect of transcriptional control that has become apparent only relatively recently is its reliance on DNA sequences spread through three-dimensional space of the nucleus. Indeed, onedimensional physical distances between genes or between genes and some of their regulatory sequences are not a hindrance for coordinated activation and repression of their activities. It is possible that different genes can extend out of their chromosome territories to access one of the shared transcription factories mentioned above (Fig. 3.12). This is very similar to the looping-out mechanisms proposed for long-range action of distal regulatory elements, such as an enhancer or locus control sequence (Fig. 3.13)234. The examination of long-range control of gene transcription has been greatly facilitated by the emergence of novel methods, such as the chromosome conformation capture, or 3C, method and the tagging and recovery of associated proteins, or RNA trap assay234. Similar to the ChIP-on-chip assay, these methods also rely on formaldehyde fixation of cells or tissues and the subsequent identification of the DNA sequence captured by the cross-linked chromatin. Using these methods, chromatin-loop interactions between enhancers and their target promoters, often tens of kilobases away, have now been conclusively demonstrated. Some enhancers have been found at distances of up to 1 Mb from their gene target, which underscores the need to limit their action (which is usually promiscuous) to their target genes only. As already mentioned, a gene or gene cluster can achieve certain autonomy through the establishment of chromatin boundaries. One way of creating such boundaries is through the action of specific DNA sequences and associated proteins, termed boundary or insulator sequences (Fig. 3.11). Such sequences are also important in
88
GENOME STRUCTURE AND FUNCTION LCR
Gene cluster
Direct regulatory interaction of LCR with genes
Fig. 3.13 Long-distance transcriptional regulation. DNA can form loops that directly juxtapose enhancer-bound regulatory proteins with promoter-bound transcription complexes. LCR, locuscontrol region.
halting the spread of transcriptionally repressive condensed chromatin structures. Long-range gene-regulatory interactions and the organization of the genome in active chromatin domains immediately explain the large variability in transgene expression levels normally observed after integration of a gene-expression construct at diverse chromosomal positions. Disruption of long-range regulatory action, for example, as a consequence of deletions or other types of genome rearrangement, can result in geneexpression alterations. Indeed, position-independence of gene expression is lost when one essential cis-regulatory region is deleted235. The possibility that this type of event contributes to increased aging-related stochasticity of gene expression will be discussed in Chapter 7. In summary, gene-transcription regulation depends on the three-dimensional interaction of a large variety of DNA sequences—which are not limited to the immediate vicinity of a gene—interacting with a multitude of protein and RNA cofactors. Transcription activation and repression through these cofactors appear to depend on higher-order DNA structure, with most of the genome in a permanently repressed, heterochromatin state. This view of gene-transcription regulation significantly differs from earlier ideas and has immediate relevance for potential mechanisms of aging-related cellular dysfunction. Indeed, the enormous advantage of having so many factors impacting on the transcriptional activity of a gene, often in parallel, is the opportunity to provide redundancy as well as fine regulation of cellular function. However, it also creates a myriad of access points to bring about slight, but cumulative, defects that may ultimately adversely affect function.
GENOME STRUCTURE AND FUNCTION
89
3.5 Conclusions Almost five decades after it was first postulated that we age because our genomes age, what can we learn from this brief overview of our most recent insights into the genome’s structure and functional organization? For decades, attention has been focused on a genome consisting primarily of protein-coding genes without seriously considering a function for the vast majority of DNA sequences in our genome, which do not encode proteins, but are often transcribed. It is only very recently that this focus has shifted to non-coding DNA sequences, RNA-coding DNA, their transcripts, and their possible role in the systems engineering process that operates our genomes as the information organelles that control development and functional maintenance in adulthood. The first lesson clearly is the realization that in this vast amount of functional primary sequence, with the extra dimensions of higher-order DNA and nuclear architecture, errors—some significant, many subtle—are inevitable. This is especially true during the perpetuation of the entire structure during cell division, including those divisions that lead to the zygote and its development into an adult. We can see the consequences of such errors already at a very early stage. In humans, chromosomal aberrations and possibly other types of mutation are responsible for the vast majority of spontaneous abortions and still births. This is in spite of the selection opportunity to eliminate genetically aberrant germ cells. The fact that most of the time adverse effects of genomic errors are not immediately obvious, at least not during development and early adulthood, does not mean there are no genomic errors or that their effects are unimportant. The difficulties in detecting small differences in organismal fitness limit the identification of subtle effects, which may be especially important in somatic cells on the long term. As we can deduce from Chapter 2, heritable deficiencies in maintaining the genome code causing harmful effects only at old age are not eliminated by natural selection. A second lesson that can be learnt from our current insights is the highly sophisticated manner with which the genome regulates its transactions, resulting in unprecedented levels of precision. Yet one cannot escape the notion that this entire symphony of cellular function is in some way inefficient, with multiple, often overlapping and haphazard, levels of control. This serves as another reminder that evolution does not work as a designer that pursues a long-term plan for the most straightforward operational system. The way we now understand Darwinian evolution gives us reason to expect complex, sometimes messy, interactions between all possible levels of genome functioning, including transcription regulation, maintenance and replication. The system, in spite of its many safeguards, including active repair, redundancy, and mechanisms for cell elimination, certainly has not been designed to last the ages. Finally, from what we have seen, the genome’s Achilles heel seems to reside in its need to provide function through transcription. Maintenance of both the DNA primary sequence and the higher-order regulatory punctuation of the genome are critically
90
GENOME STRUCTURE AND FUNCTION
important for its long-term performance in delivering the right protein at the right time in the right place. As we shall see, in spite of the sophistication of genome-control systems, it is not going to meet this requirement for very much longer than strictly necessary to perpetuate itself. Indeed, threats to its integrity come from both outside and from within, with chemical DNA damage the most prominent source of genomic errors. This becomes apparent when genetic defects, from subtle to severe, predispose the organism to making mistakes. In the next chapter I will discuss the many overlapping systems that are available to correct chemical damage to the DNA of the genome.
4
Genome maintenance
Genomes exist by virtue of their capacity to self-replicate as the key attributes for organismal perpetuation. Whereas the possibility of genetic errors has been recognized since the discovery of the DNA double helix, the magnitude of lapses in information transfer was not immediately clear. Indeed, a major hindrance in gaining a full understanding of the potential impact of genome instability on a living organism was the initial lack of insight into the mechanisms by which genomes may disintegrate over time and the factors influencing such process. Only in the early 1960s, after the discovery of excision repair in bacteria by Richard Setlow (Upton, NY, USA)236, later confirmed by Boyce and HowardFlanders237, did suspicion arise that spontaneous DNA alterations would be much more frequent, were it not for the continuous monitoring of the genome for changes in DNA chemical structure. Once the critical importance of maintaining genome integrity was realized, the concept was borne that the genes specifying genome maintenance may belong to a category of genes called longevity-assurance genes238,239. However, at the time, the enormous complexity of the various interconnected systems for detecting and repairing damage to the genome was still unknown240. Genome maintenance is a general term and includes systems for sensing and signaling the presence of DNA damage, the repair of such damage or its tolerance, as well as the proper reconstruction of DNA higher-order structure after repair is completed. It also includes systems to continuously monitor the key processes of genome information transfer, from DNA replication to mitosis, to prevent or correct errors and in a broad sense also those molecules and enzymes that act to prevent damage to the genome or buffer its adverse effects. The core of the genome is of course DNA and a major source of genome instability is the continuous introduction of lesions in this molecule by a variety of exogenous and endogenous physical, chemical, and biological agents. The removal of such lesions occurs through DNA repair and is considered so important that DNA repair and genome maintenance are often used as synonyms. I will do that too, occasionally. As we have seen in Chapter 3, DNA is a very stable molecule. However, under physiological conditions, DNA readily reacts with a variety of agents, most notably water and oxygen, to gradually disintegrate. Damage to the genome, resulting in errors in its information content, is not something that has emerged over time in modern organisms, but has from the beginning been intricately associated with life as the driver of evolutionary change (see Chapter 2). The first replicating nucleic acids, probably RNAs, evolved over 3.5 billion years ago in an environment with little molecular oxygen. The absence
92
GENOME MAINTENANCE
of an ozone layer around the primitive earth would have subjected any exposed early replicators to high fluxes of damaging ultraviolet radiation. Once initiated, life based on nucleic acids had to cope with the problem of maintaining a balance between a sufficient amount of genetic variation and its integrity as an individual entity; it had to balance evolvability against longevity. Probably the most ancient longevity genes selected must have been those involved in catalyzing the chemical reactions useful to maintaining the genomes of the individual protocells sufficiently long for them to multiply and compete with other individual protocells. Hence, some form of DNA repair was essential from the very beginning, which is reflected by the wide diversity of genome-maintenance pathways in all current species. Since all organisms are derived from a common ancestor, with new species arising by a splitting of one population into two or more populations that do not cross-breed, the evolutionary relationship of systems such as DNA repair can be studied by creating evolutionary trees of gene sequence data. As discussed in Chapter 1, such phylogenetic analysis is greatly facilitated by the availability of complete sequence information on the genomes of a great many species in all three kingdoms of life: Archaea, Bacteria, and Eukarya. The evolutionary history of DNA-repair enzymes is complex and characterized by extensive gene duplication, gene loss, and horizontal (also termed lateral) gene transfer between species. In horizontal transfer genetic information is passed on, not just to progeny but to other individuals in the same population or even to members of other species. When this occurs, the introduced gene brings into the cell its own history, which does not reflect that of the host cell. An extreme case of horizontal transfer took place about 1 billion years ago when Eubacteria entered into a symbiosis with Archaea-type host cells to become the mitochondria of eukaryotes201. This explains why some eukaryotic DNA-repair proteins can be traced to bacterial and archaeal roots. Another characteristic of the evolutionary history of DNA repair is the relative scarcity of DNA-repair proteins that are truly universal across the three kingdoms. It is possible that truly universal DNA-repair proteins—present in the last common ancestral genome—are limited to a RecA-like recombinase, a few helicases, nucleases, and ATPases241. There are profound differences in core replicative enzymes (but not the Y-family polymerases, which are bona fide homologs in bacteria and eukaryotes; see below), which is in striking contrast with the universal conservation of the translation machinery. It is possible that such differences in DNA-repair enzymes between the three kingdoms reflect early changes in both internal and external environment, requiring new DNA-repair functions. Genome maintenance may have been the first system that became subject to natural selection because of its critical importance to life as we know it242. Originally the most ancient DNA-repair systems may have functioned in protecting an RNA-based ancestral cell. The inherent instability of RNA—because of the presence of the 2'-hydroxyl group of ribose, making its phosphodiester bonds very susceptible to hydrolysis—could have been the primary reason for the transition to a DNA/protein world, which eventually
GENOME MAINTENANCE
93
permitted the evolution to complex, multicellular organisms. Genome maintenance, now to a large extent driven by DNA-damage signaling and repair processes, remained of the utmost importance as a cellular defense mechanism. This is evident from the large investment that cells make in DNA-repair enzymes, with several percent of the coding capacity of organisms devoted solely to DNA-repair functions. In some cases, one entire protein molecule is sacrificed for the removal of a single lesion. Whereas random changes occur in all biological macromolecules, either as errors in their synthesis or from the reaction with environmental or endogenous damaging agents, the consequences for the DNA of the genome are especially dire. For most biopolymers, including RNA, proteins, and lipids, the effects of damage or synthetic errors are minimized by turnover and replacement of altered molecules. DNA is distinctive in that its information content must be transmitted virtually intact from one cell to another during cell division or reproduction of an organism. Moreover, even in non-dividing cells, major portions of the genome are continuously needed as regulated templates to carry out specific functions. It is not surprising, therefore, that defects in genome-maintenance systems are incompatible with normal life. Such defects are often embryonically lethal and defective or diminished DNA-repair capacity can cause disease, including cancer, and accelerate normal symptoms of aging (see below and Chapter 5). The specific need of DNA for stability is satisfied by the various genome-maintenance pathways, which will be reviewed in this chapter with a focus on eukaryotes and mammalian cells. Similar to Chapter 3, this overview is not exhaustive and mainly serves in providing the context for the main purpose of this book: to sketch the rationale for genome alterations as a possible molecular basis of aging. For more detailed information the reader is referred to the excellent textbook, DNA Repair and Mutagenesis, by Friedberg, Walker, Siede, Wood, Schultz and Ellenberger240 and several recent reviews referred to in the text.
4.1 Why genome maintenance? The unique position of the DNA of the genome among biological macromolecules, that is, its lack of opportunities to easily replace the genetic information primarily embedded in its sequences through natural turnover like proteins and lipids, does not automatically imply the need for advanced maintenance systems. Indeed, genome integrity can be protected through redundancy, when the same function can be performed by other, identical elements, and degeneracy, when elements that are structurally different are able to perform the same function. It is also possible that large sections of the genome are dispensable, which is suggested, but not proven, by the generation of viable mice homozygous for large deletions of the non-coding DNA referred to as gene deserts243.
94
GENOME MAINTENANCE
However, structural redundancy and degeneracy of a genome are apparently not enough to survive natural levels of DNA damage. This is illustrated by the bacterium Deinococcus radiodurans. This organism is extremely resistant to ionizing radiation and many other agents that damage DNA. This is likely to be an adaptation to extreme desiccation followed by rehydration, which causes very high levels of DNA damage. This bacterium is highly polyploid (4–10 genome equivalents), which would provide it with additional gene copies to complement inactivated ones. However, this is likely not the main function of the polyploidy. Instead, this form of redundancy serves to increase the efficiency of homologous recombination (HR) repair, and not primarily to provide back-up copies of genes. In fact, apart from the polyploidy, there is also a redundancy in DNA-repair functions244. Indeed, almost all possible repair mechanisms that have been identified in prokaryotes, many of them overlapping, are encoded in the genome of D. radiodurans. The genetic cost of maintaining genome function solely by structural redundancy (creating back-up copies of all possible genetically encoded functions) is apparently too high, illustrating the magnitude of the problem. Structural redundancy can also be dangerous, for example, because it could cause undesirably high levels of gene expression or, in vertebrates, promote the activation of oncogenes when more copies of such genes are available. DNA repair is desperately needed because damage to the genome is a very frequent event, with all organisms continuously exposed to a large variety of genotoxic agents. UV radiation has already been mentioned as a major exogenous DNA-damaging agent, especially relevant during the early stages of life in the absence of an ozone layer. Modern humans are also exposed to UV radiation in the form of sunlight, but to other agents as well, including ionizing radiation and a range of genotoxic chemicals in our environment, including dietary components and air-borne pollutants. However, the main reason for all organisms to invest so heavily in DNA repair involves an enemy from within rather than from outside. All organisms dependent on water and oxygen for their existence are continuously threatened by endogenous DNA damage. It has been estimated that many thousands of spontaneous lesions are induced in the DNA of the genome of a cell each day, solely as a consequence of hydrolysis. Another major endogenous source of spontaneous DNA damage in aerobic organisms is oxidation. Oxidation, as we have seen, was first recognized by Denham Harman as a logical explanation for the various forms of cell and tissue damage observed to occur during aging23,74,245. As by-products of oxidative phosphorylation and other biological and physiological processes, oxygen radicals can inflict a variety of damages on cellular DNA and other biological macromolecules like proteins and lipids. The lesions induced in DNA by free radicals are diverse and include a variety of adducts246 as well as abasic sites, cross-links, DNA single-strand breaks (SSBs) and DNA double-strand breaks (DSBs). Figure 4.1 schematically depicts the different types of DNA chemical damage that can be induced by exogenous and endogenous agents: inter- or intra-chromosomal cross-links, DNA SSBs or DSBs, and a variety of base adducts subdivided into bulky and small adducts.
GENOME MAINTENANCE
T
A
G
C
95
Single-strand break
Inter-strand cross-link C
G
A
T
Thymine dimer
T A
C
G
G
C
Double-strand break
T Missing base G
C
C
G
A
T
Damaged base
Fig. 4.1 A simplified depiction of the spectrum of chemical damage to DNA as can be found in the living cell.
Many DNA lesions, including the ubiquitous oxidative base alterations, have been identified and measured in mammals, both upon their induction in the DNA and as adducts, often as excretion products in the urine of rodents or humans. The excretion of adducts is a result of their excision by DNA-repair systems247. Hence, spontaneous DNA damage reflects a steady-state situation and is generally not permanent.
96
GENOME MAINTENANCE
DNA damage
Apoptosis, cell senescence
Transcriptional interference
Mutation, altered chromatin
Aging
Cancer
Fig. 4.2 Adverse effects as a result of DNA damage that can contribute to aging. Note that cancer, itself an important component of aging, can be suppressed by the same cellular end points that promote degenerative senescent changes.
In principle, DNA damage can result in three main types of adverse effect (Fig. 4.2). First, DNA damage can interfere with transcription. Second, DNA damage can initiate the cellular signals that lead to programmed cell death (apoptosis) or permanent cessation of mitotic activity: replicative senescence. Third, DNA damage can have long-term effects in the form of mutations: irreversible alterations in DNA sequence information. These consequences often depend on various factors, such as the type of cell, the stage of differentiation or the stage of the cell cycle, the amount and type of DNA damage, and the genomic location of the damage, for example, transcribed or non-transcribed DNA sequences, repeat elements or unique sequences. The events that may lead from these lesions to the aging phenotype are discussed in detail in the following chapters. What is clear, however, is that without efficient systems to rapidly address genome structural deficiencies and restore the original situation, spontaneous DNA damage would soon overwhelm the cell and result in a total collapse of its functional integrity. Genome maintenance can be roughly subdivided into two components: DNA-damage signaling and DNA repair per se, including tolerance (Fig. 4.3). Whereas DNA-damage signaling is primarily involved in orchestrating a large variety of cellular responses to DNA damage, DNA repair is a straightforward damage-removal system. Both branches of genome maintenance require proteins to recognize or sense the damage, which then in turn recruit other proteins to carry out the task at hand, being the signaling of a cellular response or a given repair activity. As described in Chapter 3, many of these proteins are kept at storage sites, which may include nucleoli, telomeres, and PML bodies. Upon damage infliction these proteins undergo intranuclear relocalization and translocate to nuclear foci thought to represent sites of DNA damage and repair; such sites are sometimes called repairosomes. As yet, little is known about the exact mechanisms of damage recognition and the nature of the sensors for various types of damage. Indeed, considering the fact that a diploid genome comprises about 2 m of DNA, how is the cell alerted to the presence of
GENOME MAINTENANCE
97
DNA damage
Sensors
Signaling, e.g. ATM, p53
Cell-cycle arrest, apoptosis, cellular senescence
Chromatin remodeling, e.g. H2AX
DNA repair
Fig. 4.3 The two major branches of genome maintenance. DNA repair sensu stricto aims to restore the original situation by removing the lesion. The complex of DNA-damage signaling pathways assists in these repair activities or activates cellular responses that kill or terminate mitotic activity of a cell when it is beyond repair. The elements of the two pathways depicted on the figure are discussed later in this chapter. ATM, ataxia telangiectasia-mutated.
damage? As we will see, there are indirect ways of ‘seeing’ DNA damage, but the problem of finding a DNA lesion by a repair or signaling protein can to some extent be generalized by comparing it with the recognition of specific DNA sequences by transcription factors or restriction endonucleases. In this context a model has been developed, termed facilitated diffusion, in which the protein first binds DNA non-specifically and then slides towards its target along the DNA molecule. Already in 1965 Hanawalt and Haynes postulated that DNA-damage recognition was based on proteins that they compared to ‘closefitting sleeves’, through which the DNA was threaded to detect deviations from the formal Watson and Crick structure248. However, it is unlikely that such models reflect reality and it is now considered more likely that damage recognition involves an iterative process of transient binding and dissociation. It is possible that the repair enzymes initially recognize discontinuity of the DNA double helix due to base unstacking, kinking, or nucleotide extrusion249. For lesions such as DNA breaks the recognition process may be different and based on chromatin distortion. It is as yet unclear whether the recognition process of DNA damage is similar for DNA repair and DNA damage signaling. Whereas in DNArepair damage is recognized through specific protein–DNA interactions, the DNAdamage sensors activating cellular responses to the damage are more general since they must respond to a wide variety of damage as well as problems during replication. Overlap between these two branches of genome maintenance in terms of damage recognition is nevertheless conceivable.
98
GENOME MAINTENANCE
4.2 DNA-damage signaling and cellular responses Whereas the details of the initial stages of DNA-damage signaling are not clear, it is assumed that the presence of most lesions is detected by sensor proteins that convey the damage signals to transducers, transmitting them to numerous downstream effectors. The signaling mechanism, which depends on the type of damage and whether or not the DNA sequence is transcribed, has thus far been approached almost exclusively from the perspective of the actively proliferating cell with a strong focus on the concept of cell-cycle checkpoints, introduced by Hartwell and Weinert250. Originally, a cell-cycle checkpoint was defined as a regulatory pathway that controls the order and timing of cell-cycle transitions by checking at the beginning of each new step whether the previous one is completed. For our present discussion this basically involves an examination of genome integrity at specific points in the cell cycle. Upon sensing the damage, mechanisms are activated that arrest cell-cycle progression at checkpoints in the G1/S, intraS, or G2/M phases to allow time for repair. This somewhat narrow concept of DNA-damage checkpoints has now been extended to include the whole range of cellular responses to DNA damage, which comprise: (1) activation of checkpoints to allow time for repair; (2) recruitment of and/or participation in various damage-repair systems; (3) activation of systems for damage tolerance; (4) activation of programmed cell death or replicative senescence to eliminate the affected cell or prevent it from further replication; and (5) the activation of a host of stress-response genes, the function of which is still far from clear. Since cultured, actively proliferating cells, often in the form of bacteria, yeast, or mammalian cell lines, have become our major model systems in cell biology, a bias towards studying DNA damage and repair in cycling cells is not surprising. However, most tissues in vivo contain a very small number of cells in the S phase or no proliferating cells at all, and even stem cells, the presence of which in some normal adult tissues has now been amply confirmed, exist mainly in a quiescent state. Recently, evidence emerged for cell-cycle activation and apoptosis in postmitotic neurons upon treatment with genotoxic agents251. While intriguing, the generality of such mechanisms remains to be demonstrated. Below, I will briefly summarize our current knowledge as to the nature of the various DNA-damage-sensing and -signaling mechanisms and how they interact with the multiprotein complexes specialized in different DNA-repair pathways.
4.2.1 ATM/ATR-DEPENDENT CHECKPOINTS An important role in the damage signaling process is played by the DNA-damageinducible phosphoinositide 3-kinase-related kinase (PIKK) family. For example, the presence of the highly toxic DNA DSBs in mammalian cells is often signaled by the ataxia
GENOME MAINTENANCE
99
telangiectasia-mutated (ATM) kinase, which phosphorylates proteins that initiate cell-cycle arrest, apoptosis, and DNA repair. Humans lacking functional ATM suffer from a syndrome called ataxia telangiectasia, characterized by cerebellar neurodegeneration, immunodeficiency, extreme sensitivity to ionizing radiation, and increased susceptibility to cancer. Ataxia telangiectasia is sometimes considered as a segmental progeroid syndrome, a disease characterized by the premature appearance of multiple symptoms of aging. As we will see in Chapter 5, most of these syndromes are caused by heritable mutations in genes involved in genome maintenance, by itself evidence that genome instability is a causal factor in aging. Since DSBs are so toxic, ATM is ever present near the DNA, albeit in an inactive dimeric state or as a higher-order multimer. Inactive ATM dimers are converted into catalytically active monomers, through autophosphorylation, which leads to dimer dissociation and the initiation of cellular ATM kinase activity within minutes of exposure to an agent that induces DSBs, such as ionizing radiation. According to one model, a DSB causes alterations in chromatin structure that can affect very large regions of genomic DNA, thereby activating hundreds of ATM molecules252. However, at least in vitro, ATM activation appears to be dependent on the Mre11–Rad50–Nbs1 (MRN) complex, which may sense the DSBs, recruit ATM, and dissociate the ATM dimer253. How MRN does this is unclear but it may also involve a conformational alteration. Of note, MRN unwinds the DNA ends of a break to generate single-stranded DNA, which is essential for ATM stimulation. This is of interest since the activation of ATM- and Rad3-related (ATR) kinase also requires single-stranded DNA (see below), which is an evolutionarily conserved signal for DNA damage. Whereas the exact mechanism of ATM activation is still debated there is general consensus that, once activated, ATM activates downstream cellular targets, such as p53 and Chk2, through phosphorylation. Another PIKK, the ATR kinase, has a broader specificity and participates in responses to a wide range of DNA damage, probably exclusively during the cell cycle in response to replication blockage. Unlike ATM, ATR is essential for both embryonic development and somatic cell growth. ATR exists as a complex with ATR-interacting protein (ATRIP), which is recruited to DNA single-stranded regions coated by replication protein A (RPA)254. Most DNA-repair pathways process DNA damage through RPA–singlestranded DNA intermediates and stalled replication forks expose extended regions of RPA–single-stranded DNA. Once recruited, ATR activates other checkpoint proteins through phosphorylation, including proteins of the 9–1–1 and Rad17–replication factor C (RFC) complexes (see below) and claspin. The latter, in complex with the breast cancersusceptibility protein BRCA1, is required for the ATR/ATRIP-dependent phosphorylation of Chk1. All these checkpoint proteins must be recruited independently. Eventually, ATR-dependent phosphorylation of Chk1 and p53 helps to stabilize stalled replication forks by establishing arrest at critical cell-cycle checkpoints, including intra-S- and G2/ M-phase checkpoints, and possibly signals apoptosis.
100
GENOME MAINTENANCE
By comparing different sources of DNA damage—oxidative stress, ionizing radiation, and UV radiation—it has been demonstrated that the ATM and ATR pathways overlap. However, regardless of the genotoxic agent, ATM or ATR always phosphorylated the checkpoint proteins p53, Chk1, and Chk2. These proteins are important transducers of the damage-response signal to effector proteins, including p21 and Cdc25A, that finally result in arrest at critical cell-cycle checkpoints. Thus, for all three sources of damage, the downstream targets for phosphorylation are similar; rather, it is the requirement for ATM or ATR that is different. Taken together, these findings suggest that the ATM and ATR kinases each evolved to respond to unique forms of genotoxic stress255. While primarily regulating checkpoint activation in response to DNA DSBs, there is evidence that ATM is also directly involved in the repair of such breaks by activating component proteins of the process of non-homologous end-joining (NHEJ; see below)256. In addition, ATM may function in the maintenance of the telomeres, described in the previous chapter and further below, that cap the ends of chromosomes. It may do this through direct binding to short telomeres, phosphorylation of telomere-binding proteins, or other types of regulation of telomere proteins257 (see also Chapter 5).
4.2.2 DNA-PK Another member of the PIKK family of kinases involved in sensing and signaling of DNA damage in response to genotoxic insult is the DNA-dependent protein kinase DNA-PK, best known for its role in NHEJ. DNA-PK is composed of a large catalytic subunit, DNAPKCS, and the Ku70–Ku80 dimer with high affinity for DNA ends, which guides the kinase subunit to the site of the damage, usually a DSB. DNA-PKCS homologs are present in all vertebrates, but not in invertebrates, such as yeast, nematodes, and fruit flies, in contrast to other PIKK family members, such as ATM and ATR. Apart from its role in NHEJ, DNAPKCS acts as a genuine DNA-damage sensor, transmitting signals to p53, to regulate apoptosis and cell-cycle arrest, or mediate the recruitment of DNA-repair complexes258. There is evidence that DNA-PK activity in rodent cells is much lower than in human cells259, which could point towards some role of the protein in determining lifespan differences between species (see further below). Interestingly, a fourth member of the PIKK family is the target of rapamycin (TOR), a kinase that, as we have seen in Chapter 2, plays a role in nutrient sensing. Its downregulation extends lifespan in the fly and the worm, but it seems to play no role in DNA-damage sensing or signaling.
4.2.3 THE 9–1–1 COMPLEX Another possible DNA-damage sensor is the heterotrimeric complex of Rad9, Rad1, and Hus1, known as the 9–1–1 complex. The complex is targeted to the nucleus and to sites of damaged DNA following genotoxic stress. The complex is structurally similar
GENOME MAINTENANCE
101
to proliferating cell nuclear antigen (PCNA), the sliding DNA clamp encountered in Chapter 3 in relation to DNA replication, and it is loaded onto DNA via the RFC complex. In this case Rad17 replaces the largest subunit (subunit 1) in the RFC complex to form Rad17–RFC2–5. Rad17 is essential during mammalian development since its inactivation in the mouse is embryonically lethal260. However, it is possible that this is not because of its participation in sensing DNA damage, but a consequence of its role in the repair of damage by HR. At this stage the action of the 9–1–1 complex is still speculative, but it may work alongside ATM and ATR, which are independently recruited to the sites of DNA damage. ATM and ATR have been implicated in the efficient loading of the 9–1–1 complex through phosphorylation of the Rad17–RFC clamp loader. RPA-coated singlestranded DNA enhances Rad17 binding, resulting in the recruitment of 9–1–1 to these sites, similar to the ATRIP, which is also targeted to single-stranded DNA in an RPAdependent manner. Like ATM, there is evidence that 9–1–1 is directly involved in DNA repair261. In this respect, it is possible that the 9–1–1 complex serves as a general recruiting platform for DNA-damage checkpoint proteins as well as for DNA-repair enzymes.
4.2.4 POLY(ADP-RIBOSE) POLYMERASE-1 Poly(ADP-ribose) polymerase-1 (PARP-1) is an abundant nuclear enzyme that can be activated by ionizing radiation, alkylating agents, or oxidative stress. Like DNA-PKCS, it recognizes distortions in the DNA helical backbone; first of all DNA strand breaks, but also other structures, such as single-strand regions, four-way (Holliday) junctions, and DNA hairpins. Such binding, through its double zinc-finger DNA-binding domain, leads to PARP’s catalytic activation and the display of the function first described by Pierre Chambon (Strasbourg, France) and co-workers in 1963; that is, the posttranslational modification called poly(ADP-ribose) polymerization262. This is the most extensive posttranslational modification known. The process consumes NAD, not unlike the (NAD)dependent HDAC Sir2, discussed in Chapter 2. The enzyme targets glutamate or aspartate residues in a number of proteins, including PARP-1 itself, histones, and topoisomerases. It is possible that PARP’s action facilitates access of DNA-repair enzymes to the site of damage, for example, by the covalent modification of histones or interaction of histones with auto-modified PARP. Whereas PARP-1 accounts for most of the total cellular poly(ADP-ribose) formation, other PARPs exist, including PARP-2, which can also be activated by DNA strand breaks. Interestingly, whereas mice with either PARP-1 or PARP2 deleted from their genome are viable and fertile, the combined ablation of these genes is embryonically lethal. Upon activation, PARP-1 can signal cell death, probably after severe genotoxic stress, through the depletion of cellular NAD pools. Normally, at low levels of DNA damage, poly(ADP-ribose) polymerization is a transient modification, which is rapidly reversed by the action of poly(ADP-ribose) glycohydrolase (PARG). In this context, PARP-1 is a
102
GENOME MAINTENANCE
survival factor that has been implicated in multiple DNA-repair pathways, including base-excision repair (BER; see below), and also DSB repair. PARP-1, PARP-2, and two other PARPs, TANK1 and TANK2 (called tankyrases), also play a role in telomere maintenance (see below). As yet the mechanistic details of PARP-1’s involvement in DNA-repair pathways are unclear, but modification of chromatin structure to provide accessibility to DNA-repair proteins could be important. Results thus far indicate that PARP-1 can facilitate both the compaction and decondensation of chromatin through poly(ADP-ribose) polymerization with histone and non-histone chromosomal proteins or itself as the targets, depending on the signals available. PARP-1 is one of several genome-maintenance factors that have been linked to aging. The activity of PARP was tested in white blood cells of different mammalian species stimulated with either radiation or double-stranded oligonucleotides and was found to correlate with the lifespan of the species263,264. Although correlations can be misleading and nothing is known about PARP activity in different organs and tissues, this may point to a genuine relationship between genome-maintenance capacity and species lifespan.
4.2.5 REPLICATION AND TRANSCRIPTION AS DNA-DAMAGE SENSORS As we have seen, the sensing of DNA damage has been studied most extensively in proliferating cells, in which it is strongly associated with DNA-damage checkpoints and DNA repair. Replication is of course an ideal way of sensing DNA damage, since its machinery removes the histone protein cover from the DNA and cannot proceed immediately when it encounters DNA damage. Outside the S-phase, other ways of DNA-damage sensing must come into play. Another form of DNA processing that lends itself well to monitoring DNA integrity and to activating DNA-damage signaling, also in non-replicating cells, is transcription265. Of the three RNA polymerases, RNA polymerase II covers the largest and possibly the most relevant part of the genome; that is, transcribed protein-coding genes. DNA lesions blocking the elongation of RNA polymerase II can signal apoptosis through p53 or independent of p53266. As further described below, the process of transcription-coupled repair can prevent this situation by quickly removing the lesions blocking the polymerase.
4.2.6 DNA-MISMATCH REPAIR AS A DAMAGE SENSOR DNA-mismatch repair (MMR) is principally an enzymatic pathway for editing newly replicated DNA (see below). However, apart from correcting nucleotide mismatches as a result of replication errors, MMR also recognizes DNA lesions, most notably
GENOME MAINTENANCE
103
O6-methylguanine, the major lesion produced by methylating agents. Cells with a defect in the process that normally repairs such lesions undergo programmed cell death (apoptosis), but only in the presence of an active MMR system. When such cells are also defective for MMR they are highly resistant to killing by methylating agents. This has been explained by MMR signaling of apoptosis. In addition to apoptosis, treatment of mammalian cells with methylating agents induces cell-cycle arrest in the G2/M phase, which also is dependent on a functional MMR system. This is accompanied by the activation of ATM and ATR267. The idea is that the MMR proteins MSH2 and MSH6 would recognize the damage and directly transmit the signal to the checkpoint machinery. At this point it is not clear if the MMR-dependent apoptosis and cell-cycle arrest represent genuine DNA-damage checkpoint responses or are merely by-products of MMR’s attempt to process adducts resembling true mismatches. If the repair pathways that normally repair DNA methylation damage are inactive, the MMR proteins MSH2 and MSH6 may recognize this damage and confuse it with a mismatch. Since MMR can only excise bases from the newly synthesized strand, the lesion will remain. This will result in repeated cycles of futile repair, which will keep the methylated, template strand singlestranded for much of the time and may lead to a DSB. This may then provide the ultimate signal for apoptosis268. However, alternative models in which the MMR proteins act as direct sensors capable of signaling to the apoptotic machinery have been proposed269. Hence, it is as yet unclear whether MMR is a DNA-damage sensor or if the observed signaling merely reflects its structural limitations in damage processing.
4.2.7 ROLE OF CHROMATIN STRUCTURE IN SENSING DNA DAMAGE An efficient cellular response to DNA damage requires changes in higher-order genome structure—through nucleosome rearrangements or chromatin remodeling, as it is also called. The signals eliciting responses to DNA damage are not necessarily a detection of the primary damage but could be a response to modified chromatin structure. Similar to other nuclear processes, such as the establishment of heterochromatin or transcription, chromatin remodeling in the DNA-damage response is accomplished through modifications of histone proteins. These marks include phosphate, methyl, or acetyl groups and even small proteins, such as ubiquitin or SUMO, a small, ubiquitin-related modifier. One of the best-characterized chromatin modification events in DNA-damage responses is phosphorylation of the SQ motif found in histone H2A or the H2AX histone variant in higher eukaryotes270. This modification is an early response to the induction of DNA damage, and occurs in a wide range of eukaryotic organisms, suggesting an important conserved function. This phosphorylation can be carried out by ATM, ATR, or DNA-PKCS in a redundant, overlapping manner.
104
GENOME MAINTENANCE
H2AX activation is important for the repair of DSBs, or of SSBs converted into DSBs by replication. The key role of H2AX in genome maintenance is underscored by increased tumor susceptibility, genomic instability, and sensitivity to ionizing radiation in H2AXhaploinsufficient mice271. Foci of phosphorylated H2AX (designated H2AX) are readily formed at the site of DNA DSBs. Indeed, the formation and loss of H2AX has proved a sensitive measure for DSB formation and repair. Interestingly, H2AX foci have been found to accumulate in senescing human cell cultures and in aging mice272, contributing to the now widely held belief that cellular senescence is a DNA damage response. H2AX may exert its function as a genome-maintenance factor through chromatin remodeling. H2AX phosphorylation in mammalian cells spreads into genome regions of millions of base pairs flanking a break. This influences higher-order chromatin structure and may make the DNA more accessible to DNA-repair enzymes.
4.2.8 SUMMARY In summary, there are different overlapping signaling mechanisms for DNA damage. Their utilization depends on the type of damage, the phase of the cell cycle and, to some extent, the species. DNA-damage signaling appears to be well integrated in the actual repair processes, with some sensors or transducers being actual components of a repair pathway. Sensors may work on the basis of alterations in DNA higher-order structure, which are also part of the response to DNA damage and tightly coordinated with DNA repair to facilitate access to the lesion and subsequent restoration of chromatin structure after repair is complete. The proteins that sense or detect DNA damage may be shared between pathways that signal damage to enact cell-cycle arrest, apoptosis, or replicative senescence and those that repair the damage. There may be a branch point, with the number of lesions as the main determinant of the option that is chosen. However, it is also possible that there is a distinction between DNA-damage sensors that signal repair and those that signal to enact a cellular DNA-damage response. Also in this case, the level of DNA damage could determine the choice between these two branches of the DNA-damage response. At low levels of DNA damage, signaling proteins may not be quick enough to reach the damage before it is repaired. At high levels of damage the repair systems may be saturated and signaling proteins, such as ATR, may then be able to reach the damage and signal cell-cycle arrest, apoptosis, or cellular senescence273. These options are not mutually exclusive and DNA-damage checkpoint proteins, as we have seen, can get involved in DNA repair. It is possible that cells rarely suffer from such high levels of DNA damage that an apoptotic or senescence response is necessary and that most of the time DNA-repair systems are capable of handling the situation. This may be especially true for non-dividing cells, which probably rarely need—if ever—to take action as abruptly as cells in the middle of DNA replication.
GENOME MAINTENANCE
105
4.3 DNA-repair mechanisms Before discussing the different mechanisms for repairing damage inflicted on the genome from various endogenous and exogenous sources, it is first necessary to contemplate the options of not repairing damage at all, which is in a sense the simplest solution, especially when there is reason to assume that the damage is likely not to have major adverse consequences. Such cases exist and some of them are, somewhat paradoxically, a part of the normal DNA-damage response. They will be discussed here under the heading of DNA-repair mechanisms. Then it is important to consider the next easiest form of DNA repair, which is known as direct reversal of DNA damage. The fact that there are only very few types of damage that can be repaired through such mechanisms already suggests that the problem is very complex and that such direct solutions may work for some major lesions, but do not reflect in any way the more general strategies in genome maintenance that have evolved. Indeed, most DNA damage is repaired through excision-repair pathways. The double-helical structure of DNA lends itself well to such mechanisms because of its two complementary strands. When one strand is damaged the other one can serve as the template for restoring the original situation. This offers yet another advantage of using double-stranded DNA as the carrier of the genetic information rather than single-stranded RNA or DNA and may explain why virtually all species use the DNA double helix for this purpose. Also, the DNA bases are ideally suited for repair. They cannot be easily converted, for example, by spontaneous alkylation or deamination into other natural bases. This may explain why DNA uses thymine rather than uracil as in RNA. Indeed, the repair system would be unable to distinguish a deaminated cytosine from a uracil. Such uracils are now effectively removed by uracil DNA glycosylase. One exception involves methylated CpG sites. 5-Methylcytosine can spontaneously deaminate into thymine, resulting in a mismatch with guanine. This explains why so many spontaneous mutations are found at such sites274. They also accumulate with age in different organs and tissues of the mouse275 (see also Chapter 6). The presence of two complementary strands is especially useful for excision-repair systems, which work by cutting part of the damaged strand containing the lesion out of the DNA and subsequently pasting a replacement fragment using the non-damaged strand as the template. However, there are multiple overlapping DNA-repair systems, some of which are also capable of repairing lesions opposite each other at one site on each strand, for example, DSBs or DNA interstrand cross-links. Such lesions may be less frequent, but they are also highly toxic and present a considerable problem to the cell’s arsenal of repair tools. As discussed above, possibly all these systems interact with DNA-damage checkpoints and chromatin-remodeling systems. Figure 4.4 schematically depicts the major DNA-repair pathways in mammalian cells.
106
GENOME MAINTENANCE
Bulky adducts
Single-strand breaks, small base damage
Doublestrand breaks
Inter-strand cross links
Mismatches
T5T
Telomere attrition
A G
NER
BER
HR
NHEJ
MMR
TERT/RecQ
Nucleotide Excision Repair
Base Excision Repair
Homologous Recombination
Non-Homologous End Joining
Mismatch Repair
Telomerase/ RecQ
Fig. 4.4 Schematic depiction of the main DNA-repair pathways in mammalian cells subdivided on the basis of the specific forms of DNA damage that prompt their action. TERT, telomerase reverse transcriptase.
4.3.1 TOLERANCE OF DNA DAMAGE Upon sensing DNA damage, the easiest response for the cell is to do nothing and either tolerate the damage or, when in the process of transcription or replication, activate systems allowing it to bypass the lesion. The first option may be selected by terminally differentiated cells in mammalian organisms, such as neurons. Nouspikel and Hanawalt observed that neurons do not efficiently remove DNA damage from their genome276. In general, DNA repair appears to be less active in terminally differentiated cells than in proliferating, undifferentiated cells. For example, in my own laboratory, we showed in 1986 that senescence of rat skin fibroblasts in culture (reminiscent of terminal differentiation) is associated with an approximately 50% reduction in excision-repair-associated DNA synthesis277. Interestingly, as subsequently demonstrated in Hanawalt’s laboratory for UV-irradiated human neurons, whereas the bulk of their genome is not repaired, lesions from active genes were removed swiftly. In these cells they also observed proficient repair of lesions caused by UV irradiation in the non-transcribed, opposite strand, which is unusual (see below). They called this type of repair DAR, or differentiation-associated repair, and hypothesized that in this way neurons maintain a clean template for repairing their active genes while ignoring the remaining part of their genome. While still speculative, this strategy may be yet another example of a sound allocation of resources by the organism. Indeed, genome alterations in these cells may only start having adverse effects well after the reproductive period, in which case it is only prudent not to invest any major resources in repairing non-essential parts of the genome in such cells.
GENOME MAINTENANCE
107
Ignoring lesions also occurs in proliferating cells, both of prokaryotic and eukaryotic origin, during replication. This is called translesion DNA synthesis, a process to avoid cell death when encountering unrepaired DNA damage. For example, the major mutagenic oxidative DNA lesion, 7,8-dihydro-8-oxoguanine (8-oxoG; a modification of the C8 position of guanine with its tautomeric form called 8-hydroxyguanine), is bypassed efficiently by high-fidelity DNA polymerases. However, this frequently results in the incorporation of an adenine rather than a cytosine opposite the damaged G278. This A–8-oxoG mismatch is often not recognized by the mismatch-repair systems that normally proofread DNA, resulting in G → T transversion mutations, frequently observed in human cancers. Lesion bypass, however, can also be regulated by utilizing specific enzymes dedicated to operate with high fidelity on DNA harboring specific types of DNA damage. For example, the Rev1 protein incorporates cytosines opposite abasic sites, allowing lesion bypass in both yeast and mammalian cells. In humans, pold, encoded by the XPV gene, is able to insert the correct adenine base across a cyclobutane thymidine dimer, the main UVinduced lesion. When encountering replication-blocking lesions, the use of such enzymes, often referred to as Y-family polymerases279, increases the chance of survival while maintaining a reasonable level of genome integrity. This is illustrated by the increased mutagenesis and reduced survival of cells from patients with human xeroderma pigmentosum (XP) harboring a genetic defect in the XPV gene280. The mechanism of translesion synthesis is not as straightforward as one might expect. Replication of the lagging strand is discontinuous and a lesion will not compromise replication-fork progression. However, if a replication block is located in the leading-strand template, the polymerases can become uncoupled and continued template unwinding allows nascent lagging-strand synthesis to proceed past the blocked nascent leading strand (see Chapter 3). A possible scenario for translesion synthesis in bacteria, derived in part from reconstitution experiments in vitro281, is as follows (R. Fuchs, personal communication). The lesion is bypassed to reinitiate leading-strand synthesis downstream of the lesion, which leaves a gap. Such a gap, which can be 1000 nucleotides long, can then be filled by the specialized Y polymerases in an error-prone manner. However, an alternative way of gap filling was already described by Rupp and Howard-Flanders in UV-irradiated bacteria lacking nucleotide-excision repair (NER)282. Such so-called postreplication repair involves template switching—using the newly synthesized complementary strand as a template rather than the damaged one—and recombinational gap filling283. This latter, recombinational strategy is a genuine mode of DNA repair, most notably in the repair of DSBs, and will be discussed further below. In yeast, the lesion-bypass pathway is part of the RAD6 pathway, which contributes to this organism’s resistance to DNA damage. This class of mechanisms for tolerance to DNA damage should not be confused with another form of postreplication repair, MMR, which will be discussed below. The story of the Y-family polymerases, the so-called mutator polymerases, really begins with the discovery of SOS repair in bacteria. In the SOS response about 30 genes
108
GENOME MAINTENANCE
are coordinately induced as a consequence of the proteolytic inactivation of LexA, which is greatly enhanced by the RecA protein bound to single-stranded DNA. The latter explains how the SOS response is induced by DNA-damaging agents. However, LexA can also be inactivated or its cellular concentration decreased under stressful conditions, such as starvation and bacterial senescence (elicited by starvation-induced arrest of proliferation). SOS induction and mutagenesis have been observed in aging E. coli colonies, in the absence of exogenous sources of DNA damage284. The rationale for such a survival strategy of unicellular organisms under adverse conditions has been discussed in Chapter 2: a transient increase in mutation rate provides some members of the population with a new genotype that allows them to survive and proliferate. Such adaptive mutagenesis can even take the appearance of being directed to specific genes. This phenomenon was first noted by John Cairns and Patricia Foster (then both in Boston, MA, USA) as a high frequency of mutations reversing a mutationally inactivated lacZ gene in stationary cultures of E. coli285. LacZ encodes the enzyme -galactosidase, permitting the use of lactose as a nutrient. The mutations in these stationary cultures appeared to be directed by the selection process, since they only occur after these cells had been exposed to lactose. Rather than indicating some hidden causality, however, these mutants are singled out since only they, and no others, allow the cell to resume growth. It is important to realize that a similar strategy in multicellular organisms is not very useful. Whereas there may be evolutionary advantages for all species in increasing the amount of genetic variation in the germ line (temporarily, because the fitness of such mutators rapidly declines as a consequence of Muller’s ratchet; see Chapter 2), random mutations in somatic cells offer no selective advantage, but contribute to disease and aging, the topic of this book. An exception is the highly regulated process of somatic hypermutation of immunoglobulin genes, which is critical for unfolding the full power of the immune response. In this form of adaptive mutagenesis mutations are targeted to the variable regions of rearranged immunoglobulin (Ig) genes to generate a wide variety of memory cells with high affinity for a specific antigen. The enzyme activation-induced cytosine deaminase (AID) plays a major role in this process by deaminating cytosines in Ig genes, leading to the formation of uracil. It is likely that the subsequent mutations are then caused by Y-family DNA polymerases that bypass these lesions286. There is evidence that the same enzyme also deaminates mRNA in a process called editing. RNA editing is the co- or posttranscriptional modification of the primary sequence of RNA through nucleotide deletion, insertion, or base-modification mechanisms. In mammals editing seems to exclusively use base modification. RNA-editing processes are known to create diversity in proteins involved in various pathways like lipid transport and metabolism287. For example, deamination of cytosine generates a shorter form of the apolipoprotein-B mRNA in mammalian intestine and liver. The genes encoding the Y-family polymerases induced by the SOS response in bacteria—pol IV and pol V—led to the discovery, by searching genome databases, of multiple orthologs and paralogs in eukaryotes, including mammals. All these DNA
GENOME MAINTENANCE
109
polymerases display increased error rates on undamaged DNA (as compared to replicative polymerases) and are capable of translesion synthesis of damaged DNA. In contrast to the situation in bacteria, these enzymes are constitutively present in mammalian cells and tissues and highly regulated to avoid their access to undamaged DNA280. The mechanisms to ignore DNA damage or bypass lesions at the risk of mutation are highly relevant for late-life genomic integrity. Ignoring DNA damage in postmitotic cells leads to an increased DNA damage load at old age, which may contribute to cellular degeneration and death. Likewise, mutations as a consequence of lesion bypass will accumulate with age and can easily lead to increased risk of cancer and other adverse effects. This will be further discussed in Chapter 6. However, DNA-damage tolerance can also have immediate consequences for the cell at the level of gene transcription. Indeed, translesion synthesis not only occurs during replication, but also during transcription, which has been observed in both bacteria and mammalian cells, and shown to lead to mutant transcripts (transcriptional mutagenesis)288. Since quiescent, non-replicating cells are the norm in tissues of multicellular organisms, this was an important discovery, the consequences of which are still not entirely clear. Indeed, whereas replication errors in certain genes can conceivably lead to cancer, a clonal disease, the effect of transcriptional errors may be transient and have a limited impact on an organism. Nevertheless, it is easy to see that whereas the effects of infrequent events, such as occasional misincorporation of amino acids or ribonucleotides, are likely to be limited, consistent bypassing of a persistent lesion in a gene could alter the entire transcriptional output of that gene in a given cell. Most studies on transcriptional bypass have been done in bacteria. The results indicate that one of the most frequent base changes—deaminated cytosine or uracil—is readily bypassed by E. coli RNA polymerase, resulting in mutated transcript. Fortunately, uracil is normally rapidly removed by BER, which is also the case in mammalian cells (see further below). In mammals, altered proteins as a result of transcriptional mutagenesis have been demonstrated in the brains of patients with Alzheimer’s disease and aged control individuals without dementia, but not in the brains of young control subjects289. This increase in altered protein with age may remind the reader of Leslie Orgel’s error catastrophe hypothesis (Chapter 1), which was discarded years ago, mainly because of a lack of evidence for a high frequency of altered proteins with age. To implicate transcriptional mutagenesis as a causal factor in aging it would be important to study transcriptional infidelity in the synthesis of components of the transcriptional machinery itself, which would create the positive-feedback loop originally suggested by Orgel290.
4.3.2 DIRECT REVERSAL Apart from ignoring it altogether, the next simplest way of dealing with DNA damage is its direct reversal. The main reason why this is not utilized more generally is that it involves the need for too many specialized enzymes, which is genetically costly when dealing with
110
GENOME MAINTENANCE
hundreds and perhaps thousands of different types of chemical damage in DNA. Nevertheless, three major systems of direct reversal exist: enzymatic photoreactivation of UV-induced pyrimidine dimers, reversal of alkylation damage, and ligation of DNA SSBs. In photoreactivation, the nature of which was first described by Jane and Dick Setlow291, a single enzyme, photolyase, is able to repair UV-induced DNA lesions—cyclobutane pyrimidine dimers (CPDs) or pyrimidine 6–4 pyrimidone photoproducts (6–4PPs)—by effectively reversing their formation using blue light292. Based on what has been mentioned at the beginning of this chapter regarding UV radiation as perhaps the main source of DNA damage early in the history of life, it is not surprising that such a highly specific mechanism as enzymatic photoreactivation for the speedy direct repair of the main UV-induced lesions has evolved. The system also occurs in yeast, but most likely not in placental mammals, which may have lost it during evolution. It is possible that the need for such a specialized mechanism has become much less in multicellular organisms, in which only a fraction of the cells—and then not even in all organisms—is regularly exposed to UV. In mammalian cells UV damage is generally repaired through the NER pathway (see below), which in humans is especially important for protecting the skin. Interestingly, transgenic mice have been generated expressing marsupial CPD-photolyase, which dramatically reduced acute UV effects like erythema (sunburn), hyperplasia, and apoptosis293. The direct reversal of alkylation damage in human cells can be carried out by three proteins. The O6-alkylguanine alkyl transferase (encoded by the MGMT gene in humans) catalyses the transfer of alkyl groups, varying from methyl to benzyl, from the O6 of guanine to one of the repair enzyme’s own cysteine groups. This protein is itself expended in the reaction. Less widespread direct repair enzymes in mammals are the ABH2 and ABH3 proteins. Orthologs of E. coli AlkB, these demethylases catalyze the oxidation and release of the methyl group from 1-methyladenine and 3-methylcytosine. Interestingly, ABH3 can also repair RNA294, which is in fact the main target for these lesions. O6-Alkylguanines are lesions of considerable biological importance in mammals. They are highly mutagenic and may occur spontaneously at a high rate, possibly through intracellular methyl donors, such as S-adenosyl-L-methionine. Mice lacking MGMT are highly susceptible to tumorigenesis by alkylating agents295. Interestingly, transgenic C3HeB male mice overexpressing the human MGMT gene in liver and brain were found to have significantly lower levels of hepatocellular carcinoma, a frequent spontaneous tumor in this mouse strain296. This indicates that O6-alkylguanines occur spontaneously and contribute significantly to the risk of liver cancer in animals of this strain. E. coli and other prokaryotes are able to rapidly increase the amount of their O6methylguanine-DNA methyltransferase (O6-MGT I) activity upon chronic exposure to alkylation stress297. Discovered by Leona Samson (Cambridge, MA, USA) in the laboratory of John Cairns, then in Boston, MA, USA, this response is called the adaptive response, which can lead from the 1–2 molecules of the protein that are normally present to several thousands of such molecules in adapted cells. There is no evidence for such a
GENOME MAINTENANCE
111
response by the mammalian MGMT gene, which in this respect resembles the E. coli O6-MGT II gene, which is also not inducible (at least, not in an O6-MGT I mutant). DNA SSBs can be repaired through direct reversal, simply by rejoining the ends using ligase I. However, SSBs, for example, as inflicted by ionizing radiation or ROS, possess abnormal 3' and 5' termini, which need to be restored to proper 3'-hydroxyl and 5'-phosphate moieties. Therefore, most SSBs are repaired through the BER process (see below).
4.3.3 DNA-EXCISION REPAIR There are three excision-repair pathways: BER, NER, and MMR. They all employ the double-helical nature of the DNA to remove lesions from one strand and restore the original situation using the second strand as a template (Fig. 4.5). Base-excision repair
DNA glycosylase
DNA-damage detection complex
Recognition by DNA polymerase * *
Preincision/helicase complex
GATC GA A T TC CTAGC CT T A AG
DNA polymerase
GATCG GA A T TC CTAGC CT G C AA
GATCG GA A TT C CTAGC CT G TA A
GATC A A CG CTAGC T T GCT TA A
AP endonuclease
Mismatch repair
Nucleotide-excision repair
G A T C G* GA A T TC CTAGC CT T A AG
* Excision
Excision *
GATCG GA A T TC CTAGC CT T A AG
Resynthesis DNA ligase GATCG GA A T TC CTAGC CT T A AG
DNA polymerase
DNA ligase GATCG GA A T TC CTAGC CT T A AG
Fig. 4.5 Simplified version of the three excision-repair pathways: base-excision repair (BER), nucleotide-excision repair (NER), and DNA mismatch repair (MMR). They all repair DNA damage through excision and resynthesis using the opposite strand as a template. AP, abasic (see text for an explanation).
112
GENOME MAINTENANCE
4.3.3.1 Base excision repair (BER) BER is specifically devoted to the repair of small base damage, SSBs, and abasic sites, as a consequence of such processes as oxidation and non-enzymatic hydrolysis of base–sugar bonds. These lesions occur at high frequency; there are tens of thousands of lesions per day in a typical cell. Spontaneous base damage can occur, for example, as a consequence of hydrolytic deamination of cytosine, resulting in uracil. Reactive oxygen species (ROS) can cause a large number of different types of small base damage, including the 8-oxoG lesion mentioned above and thymine glycol298. To repair the large variety of small base alterations, BER relies on a battery of enzymes, termed glycosylases, that can recognize a specific type of damaged base and cleave the N-glycosylic bond between the base and the sugar. There are glycosylases specific for deaminated bases, alkylated bases, oxidized bases, and base mismatches. Examples are uracil-DNA glycosylase (UNG), recognizing deaminated cytosine (uracil), and TDG and MBD4 DNA glycosylases, excising thymine when it is mispaired with guanine as a consequence of deamination of 5-methylcytosine. There are many others, partially overlapping in their activities and also including a number of enzymes that recognize oxidative lesions, such as OGG1 DNA glycosylase excising 8-oxoG from DNA (its ortholog in bacteria is Fpg), and NTH, removing oxidized pyrimidines. When occurring in transcribed sequences, oxidative base damage can also be repaired by transcription-coupled repair (see below). The repair of endogenous DNA base damage was reviewed extensively by Barnes and Lindahl in 2004298. Removal of damaged bases by hydrolytic glycosylases leaves abasic sites, or AP sites, which can also be a direct product of damage. Once initiated, BER can proceed along two mechanistically different lines: short- or long-patch BER. The distinction involves the size of the repair patch, the number of nucleotides replaced. For short-patch BER, which is the major pathway, this is only a single nucleotide. Long-patch BER involves between two and 10 nucleotides. The enzyme that recognizes and cleaves the 5' phosphodiester bond at AP sites is AP endonuclease (APE1), the key component of BER after the glycosylase has done its work. There is another enzyme that can cleave at the AP site, this time at its 3 site. This enzyme, designated AP lyase, is associated with some glycosylases as well as with the main polymerase involved in BER, pol . The exact mechanistic details as to how the cell decides which activity and which branch of the BER pathway to use for which lesion, is not yet entirely clear and a discussion is beyond the scope of this book. The reader is referred to several excellent recent reviews283,298,299. Briefly, in the short-patch pathway pol is recruited to fill the one-nucleotide gap. Ligation occurs through the Lig3–XRCC1 complex. In the less utilized long-patch sub-pathway, the replicative DNA polymerases pol/ or pol , in a complex with RFC/PCNA, displace the strand, generating a flap of between two and 10 nucleotides, which is subsequently cut off by FEN1 endonuclease. The nick is closed by ligase I. As already mentioned, SSBs, such as those induced by ionizing radiation or ROS, are also repaired through the BER pathway. For this purpose, the enzyme polynucleotide kinase (PNK) or
GENOME MAINTENANCE
113
APE1, in a complex with XRCC1, converts the damaged termini into 5'-phosphate and 3'-hydroxyl moieties300. Apart from the core players, accessory factors likely play a role in BER. As mentioned above, the ring-like PCNA DNA clamp in combination with the RFC clamp loader organizes various proteins involved in DNA replication, DNA repair, DNA modification, and chromatin remodeling. As such, it stimulates FEN1 cleavage of the flap in long-patch BER. A similar role can be played by the DNA-damage sensor complex 9–1–1, which is structurally and functionally similar to PCNA (see above). As already mentioned, members of the PARP family may be involved in BER by binding to SSBs, either those arising during BER or breaks that are induced by ROS or ionizing radiation. PARP is then automodified, as described above. Since BER proceeds efficiently in extracts of Parp-1-null cells, such a role is not essential. In this respect it has been postulated that PARP is merely facilitating access to the DNA through local chromatin relaxation. As it turned out, BER is not only important in the maintenance of the nuclear genome, but also highly proficient in the repair of small base damage in mitochondrial DNA (mtDNA)301. This is not surprising since mitochondria are the main sites of ROS production, and ROS in turn cause the lesions that are the main substrate for BER, as we have seen. Only the shortpatch variant of BER appears to be active in repairing lesions in mtDNA.As discussed further below there is some evidence that this activity declines with age in rodents. BER must be a critically important genome-maintenance system, since complete absence of one of its core components is incompatible with life. For example, the homozygous deletion of APE1, XRCC1, ligase I, or pol in the mouse germ line is embryonically lethal. Indeed, in contrast to other repair systems, there are no known human diseases caused by a complete defect in BER, although it remains possible that slight deficiencies are associated with disease or accelerated aging. On the other hand, elimination of one of its glycosylase functions has only a limited impact. For example, mice with an engineered defect in uracilDNA glycosylase have an increased level of uracil in their genome, but do not exhibit a greatly increased frequency of spontaneous mutation. They also do not display any signs of premature aging. This lack of a major phenotype may be due to the existence of a compensatory uracil-DNA glycosylase activity in the form of SMUG uracil-DNA glycosylase298. Targeted inactivation of the OGG1 DNA glycosylase in the mouse led to the accumulation of 8-oxoG and a modest increase in spontaneous mutation frequency in non-proliferative tissues, but no increased spontaneous tumors or other marked pathological changes302. The lifespan of these animals was not significantly shorter and there were no signs of premature aging. The nature of a possible backup activity for this enzyme is not known, but transcription-coupled repair or NER may contribute. The results with these and other mouse glycosylase mutants suggest that there is significant overlap between glycosylases in recognizing different lesions as well as overlap with other repair pathways. This overlap may to some extent compensate for the inherent inefficiency of BER in needing so many different glycosylases to accommodate the
114
GENOME MAINTENANCE
wide range of DNA lesions in a cell. A battery of different enzymes may be unavoidable because it may simply not be possible to design a general recognition system for all possible types of small base damage, which is in striking contrast to the next system to be discussed, NER.
4.3.3.2 Nucleotide excision repair (NER) NER, a general repair system that removes damaged DNA bases by dual incision of the affected strand followed by resynthesis of 24–32 bp (in eukaryotes), is inextricably linked to the heritable human disease xeroderma pigmentosum (XP). XP patients develop fatal skin cancers when exposed to sunlight and in 1967 it occurred to James Cleaver (San Francisco, CA, USA), then a postdoctoral fellow at the laboratory of Bob Painter at the University of California at San Francisco, that the high susceptibility to cancer in these patients could be related to failure of DNA repair of sunlight-induced DNA lesions303. At the time, DNA repair had just been discovered, in bacteria (see the beginning of this chapter), and the first methods to measure the repair of UV-induced damage had been developed. Pettijohn and Hanawalt were the first to describe the patching step in the process by showing after irradiation that bromouracil incorporation occurred in small patches and did not give the density shifts expected from semiconservative DNA replication (see the description of the Meselson and Stahl experiment in Chapter 1). They called this repair replication304. Painter himself had developed the techniques of autoradiographic and equilibrium density-gradient detection of repair (unscheduled DNA synthesis and repair replication)305. Unscheduled DNA synthesis is a dramatically visual method of measuring NER, which has a far larger patch size than BER (Fig. 4.6). Using these methods Cleaver found that XP cells were indeed repair-deficient306. The next key player in this field was Dirk Bootsma of the Erasmus University in Rotterdam, the Netherlands, who fused cells from different XP patients, to see if they were able to complement each other by restoring UV-induced unscheduled DNA synthesis in the heterodikaryons. If so, then the genetic defect in DNA repair in each of the two complementing cells was in different genetic complementation groups required for excision repair307. Soon, this kind of somatic-cell genetic analysis using cells of XP patients from different families indicated multiple genetic complementation groups for XP, which were designated XPA–XPG. These observations, in turn, suggested that multiple genes are required for the particular mode of excision repair demonstrated in normal human cells and that mutational inactivation of any one of these genes results in XP. NER is essentially different from BER in the sense that it has no battery of enzymes to recognize specific chemical groups that make up the lesion, but instead recognizes major backbone distortions of the DNA. Such bulky lesions include the aforementioned UV-induced lesions, especially 6–4PPs, and damage created by electrophilic metabolites of carcinogenic polycyclic hydrocarbons, formed during inefficient combustion of fossil
GENOME MAINTENANCE
115
Fig. 4.6 UV-induced unscheduled DNA synthesis in rat fibroblasts. After UV irradiation the incorporation of [3H]thymidine is visualized by autoradiography. The occasional S-phase cell is indicated by a completely black nucleus. The significant, but much lower levels of [3H]thymidine incorporation, revealed by the black grains above the nuclei, indicate excision repair of the UV-induced lesions.
fuels and found in cigarette smoke, car exhaust fumes, and barbecued meat. The best known example is benzo(a)pyrene, which has been studied extensively as a genotoxic agent (Chapter 6). NER has two sub-pathways, differing in damage recognition but sharing the same downstream repair machinery: global genome NER (GG-NER) for the removal of distorting lesions anywhere in the genome and transcription-coupled NER (TC-NER) for the elimination of distorting DNA damage blocking transcription. As already briefly discussed above, in TC-NER lesions are sensed by RNA polymerase II itself and the resulting signals can activate either repair or apoptosis. As the name suggests, TC-NER involves the removal of lesions from the transcribed strand in active genes, which for some lesions, such as CPDs, is much faster than GG-NER. The pathway was discovered by Vilhelm Bohr (Baltimore, MD, USA), then in the laboratory of Philip Hanawalt (Stanford, CA, USA)308. Interestingly, it is now clear that TC-NER also functions in removing non-NER lesions. Whereas GG-NER can also remove some types of oxidative lesion, TC-NER acts on those oxidative lesions thought to be the exclusive domain of BER309.Whereas this is controversial
116
GENOME MAINTENANCE
in view of the recent retraction of several major publications310, it remains a likely possibility. In this sense, TC-NER is now to some extent recognized as a repair pathway in its own right and often just referred to as transcription-coupled repair. In TC-NER, DNA-damage sensing occurs by RNA polymerase II, thereby arresting transcription. Transcription arrest is followed by the recruitment of the products of the CSA and CSB genes, with the CSB gene product possibly acting in a complex with the basal transcription factor IIH TFIIH, and perhaps the XPG protein. TFIIH, also needed for transcription initiation (see Chapter 3), has then already dissociated and needs to be recruited again. Its XPB and XPD helicase protein subunits are again needed to open up the helix, in this case a 24–32-nucleotide stretch containing the damaged region. The details of the early steps in TC-NER are not yet known, but the CSA and CSB proteins are thought to be involved in displacing RNA polymerase stalled by a DNA lesion and recruiting the NER (and perhaps BER) machinery to the site of the lesion. The CSB protein is a member of the SWI/SNF family of ATP-dependent chromatin-remodeling factors (see Chapter 3) and changing DNA higher-order structure may be an important component of its role in TC-NER. In addition, various subunits of TFIIH and also the XPG nuclease may assist in the removal of the RNA polymerase and RNA. At this point TC-NER merges with GG-NER. Of note, heritable mutations in CSA or CSB cause Cockayne syndrome, an autosomal recessive disorder characterized by progressive postnatal growth failure, neurological dysfunction, and a short lifespan of about 12 years on average. This disease, which shows no signs of increased cancer, is often considered as a segmental progeroid syndrome and will be discussed in more detail in Chapter 5. It is only one of several human disorders associated with NER genetic defects, underscoring the importance of the pathway for human health, but also pointing towards a certain tolerance for its absence, which is in striking contrast with BER. In GG-NER the situation with respect to DNA-damage sensing is far more complex than for TC-NER. As we have seen, TC-NER is able to quickly recognize lesions, including those that are normally a substrate for BER, on the basis of transcription arrest. In this respect, TC-NER functions, to some extent, as a backup system for BER, by acting on non-bulky lesions, such as oxidized or methylated bases. However, certain helix-distorting oxidative lesions, such as cyclopurines, fall within the realm of GG-NER. It has been speculated that accumulation of these lesions in the brain may explain the observed progressive atrophy of the cerebral cortex of XP patients311. Because its recognition system is based on distortions of the DNA double helix, GGNER may also occasionally attack natural configurations of undamaged DNA, evidence for which has indeed been obtained using cell-free extracts acting on synthetic DNA fragments312. These findings suggest that even in non-dividing cells there can be considerable DNA turnover due to this kind of gratuitous repair, which may then contribute to mutation accumulation during aging, due to errors during resynthesis. However, this is merely
GENOME MAINTENANCE
117
speculation because GG-NER is likely to abort when no lesion is present, based on the damage-verification role of XPA (see below). A key role in lesion recognition in GG-NER in the context of chromatin organization is probably played by the damaged-DNA-binding protein (DDB), at least in the case of UV-induced pyrimidine dimers. This protein may activate the DNA-damage checkpoint protein ATR and also initiates GG-NER, by bending the DNA to the lesion, so as to allow XPC to bind to it. The important role of DDB in NER in a situation in vivo is indicated by the fact that a heritable mutation in the gene XPE, encoding its small, p48 subunit (DDB2), can cause XP. The gene encoding p48 is regulated by p53313 (see also below). Somewhat similar to the role of PARP in BER, DDB is not required for human NER in vitro, as became evident in a reconstitution experiment. Similar to PARP, it has therefore been speculated that DDB is especially important in facilitating repair in a chromatin context. For example, a role of DDB in nucleosome unfolding may be the key in providing access to XPC and XPA and other components of the NER machinery to damaged DNA. It is now generally assumed that GG-NER begins with the binding at the site of the lesion of a heterodimer consisting of the XPC and HR23B proteins, facilitated as we have seen by DDB (at least, for pyrimidine dimers). Binding of XPC/HR23B and, possibly, its action in increasing the DNA backbone distortion, permits the binding of TFIIH, which plays a very similar role as in TC-NER; opening up the helix through the helicase activity of two of its 10 subunits, XPB and XPD. Here again we should pause to mention the existence of human heritable diseases caused by mutations in some of these genes. Mutations in XPC or XPD can result in XP, whereas mutations in XPB cause XPCS, a combination of XP and Cockayne syndrome. Interestingly, different mutations in the same gene can lead to different diseases. An extreme example is XPD, mutations in which can not only lead to XP, but also to Cockayne syndrome or trichothiodystrophy. This strong, complicated disease link of NER explains much of its enormous appeal as a subject of study, to both basic scientists and clinicians. A detailed discussion of NER-associated diseases is beyond the scope of this book and the reader is referred to recent reviews314,315. However, the subject will come back repeatedly. Opening up the DNA helix by the action of TFIIH is followed by the recruitment of XPA, RPA, and XPG; XPC and HR23B are released. RPA binds to both strands and keeps them apart. XPA probably acts in damage verification and in guiding subsequent cleavage of the damaged strand. At this stage GG-NER has come together with TC-NER. Hence, the difference between the two sub-pathways is really in the damage-recognition steps. Therefore, defects at this and the next stages affect NER as a whole. For example, whereas defects in XPC affect only GG-NER and defects in CSA or CSB only TC-NER, XPA inactivation silences the entire NER pathway. However, CSB (and possibly CSA) may also be involved in general transcription316 and could be directly linked to BER for the repair of oxidative damage317, which may explain why Cockayne syndrome is so different from XP.
118
GENOME MAINTENANCE
The downstream part of NER involves the incision at both sides of the lesion, at the 3' site by XPG and at the 5' site by a dimer composed of XPF and ERCC1. The distance between the incisions is about 30 bases, but can vary with each repair event. The ERCC1– XPF endonuclease is also essential for interstrand cross-link repair (see further below). It should be noted that ERCC1 represents the formal nomenclature of the NER genes; that is, excision repair cross complementing. The reason that it has no XP-derived name is the absence of a patient with a mutation in this gene, which is in contrast to its sister subunit XPF and the other endonuclease XPG. After dual incision, the gap is filled in by DNA synthesis, using the opposite, normal DNA strand as a template, by DNA polymerases and , followed by ligation. NER in E coli is very similar to NER in mammals, but simpler. However, it should be noted that with respect to UV damage, bacteria, bacteriophages, and some eukaryotic viruses contain up to three distinct mechanisms to initiate the repair of UV-induced dipyrimidine adducts: NER, BER, and photoreversal. Photoreversal has been discussed. To initiate BER, glycosylases, such as T4 pyrimidine dimer glycosylase or the Micrococcus luteus UV endonuclease, are necessary240. Similar to the situation in BER, genetically modified mice are now playing a major role in unraveling the mechanisms of NER and the pathophysiological consequences of deficiencies in its performance. One of the first of these mouse models for NER was a knockout of the gene XPA, made independently in the laboratories of Harry van Steeg (Bilthoven, the Netherlands)318 and Kiyoji Tanaka (Osaka, Japan)319. Similar to their human counterparts, these mice were highly sensitive to genotoxic agents inducing bulky adducts, such as UV or benzo(a)pyrene. By now, almost all gene defects involved in NERrelated diseases have been modeled in the mouse, most of them in the laboratory of Jan Hoeijmakers (Rotterdam, the Netherlands). Hoeijmakers and collaborators were able to model, in the mouse, many of the exact same mutations in NER genes that caused these different diseases in humans, thereby mimicking their human phenotypes to an often remarkable extent. Interestingly, it was noted in the course of this work that several of these mouse mutants displayed multiple symptoms of premature aging, a phenomenon extensively discussed in Chapter 5. Taken together, there can be no doubt about the importance of NER as a major repair pathway, especially in humans. However, the fact that elimination of this pathway is not embryonically lethal (except when the gene is also involved in some other basic function, such as general transcription) indicates that NER is not as important as BER. This may be especially true in the mouse, since in the above-mentioned mutant mouse models symptoms were often considerably less severe than in their human counterparts. There are two possible explanations for this. First, it is conceivable that NER in humans is mainly important as a defense against UV. Since rodents are nocturnal animals and have a fur, they normally do not encounter any UV damage. This is illustrated by the aforementioned XPA-knockout mouse, which only showed symptoms when its (shaved) skin is
GENOME MAINTENANCE
119
challenged with UV or when it is fed compounds that induce bulky adducts, such as benzo(a)pyrene. Second, it is possible that NER gains in importance at older ages, for example, in a possible backup role for BER. Rodents generally have a short lifespan and may need NER to a much lesser extent than humans. Indeed, there are now a number of cases in which the consequences of a DNA-repair defect appear to be much less severe in mice than in humans. This will be further discussed in Chapter 5. The argument that NER is less important for rodents than humans can be illustrated by the so-called rodent repairadox, which has direct relevance for the relationship between DNA repair and aging320. Shortly after the clinical importance of NER was realized—with its absence in cells from XP patients so dramatically illustrated using the new autoradiographic assay—the wide diversity of DNA-repair processes was not immediately known. Indeed, UV-induced unscheduled DNA synthesis in cultured skin cells was often considered as a measure of the total DNA-repair capacity of the organism. Subsequent studies by Ronald Hart in the laboratory of Dick Setlow revealed a roughly linear correlation between UV-induced unscheduled DNA synthesis in skin fibroblasts cultured from different species—shrew, mouse, rat, hamster, cow, elephant, and human—and the lifespan of the species (plotted on a logarithmic scale)238. Most striking of these results was the dramatic difference between human and rodent cells. Indeed, around this time data from a number of laboratories indicated the almost complete lack of UV-induced CPD removal from rodent cells, an activity clearly present in human cells under the same conditions321,322. Although also in human cells the repair of CPDs is slow (about 80% in 24 h in cultured fibroblasts), the almost complete absence of this activity in rodent cells was considered strange in view of the fact that the survival rate of human and rodent cells upon UV irradiation is very similar322. The difference in CPD repair between rodent and human cells is likely due to a deficiency in rodent cells of the stimulation of NER by p53. The p53 tumor suppressor, an important mediator of cellular responses to DNA damage, affects the efficiency of NER by transcriptionally regulating the expression of the DDB2 gene (encoding the p48 protein) and the XPC gene, two important components of the NER pathway involved in DNA damage recognition. Whereas heritable mutations in the human DDB2 gene generate the E subgroup of XP (XPE), in rodent cells this gene is not induced by p53. The absence of a fully functional DDB causes a deficiency in global genome repair of CPDs (as well as 6–4PPs and possibly other types of lesion). By contrast, TC-NER in rodent cells is fully functional and it is likely that this proficiency in repairing UV-induced pyrimidine dimers in transcribed DNA strands explains why the survival of rodent cells after UV irradiation is not different from that of human cells. Indeed, arrested RNA polymerase II at a lesion is likely to induce apoptosis unless repair is swift266. Based on the above, one would expect that defects in TC-NER alone would lead to increased levels of apoptosis, thereby eliminating cells with a severely damaged genome
120
GENOME MAINTENANCE
and reducing cancer risk. On the other hand, defects in GG-NER alone may lead to increased genome instability but not to elevated levels of apoptotic cells. This is in keeping with results obtained after comparing mice with defects in XPC (no GG-NER), CSB (no TC-NER), and XPA (no NER) for signs of increased genome instability or apoptosis. XPC-deficient mice were shown to rapidly accumulate somatic mutations at the Hprt locus323 in T lymphocytes as well as in liver, spleen, kidney and lung at a lacZ transgene locus (S. Wijnhoven, personal communication). These results are in keeping with the observation that XPC-deficient mice develop multiple spontaneous lung tumors324. CSB-null mice displayed no such increased genomic instability at a lacZ transgene locus325, but instead increased levels of apoptotic cells (Y. Suh, personal communication). In XPA-deficient mice both an increased level of genomic instability in liver and kidney325 and increased numbers of apoptotic cells (Y. Suh, personal communication) were found. Whereas in the XPA mutant the absence of TC-NER would cause apoptosis, as in the CSB-deficient mouse, this may not be enough to entirely eliminate cells with accumulated damage due to the defect in GG-NER, which is also present in this mouse model. At the time the Hart and Setlow paper appeared, the mechanistic details discussed above were not known. It is possible that the low level of UV-induced GG-NER in rodents simply reflects an adaptation to nocturnal life. Interestingly, Kato et al.326, who greatly expanded the Hart and Setlow study, observed very low UV-induced repair synthesis in cells from bats, nocturnal animals that are long-lived. Of note, unscheduled DNA synthesis as a measure for NER can be influenced by numerous confounding factors, such as cell geometry, nucleoside pool sizes, and [3H]thymidine uptake. Based on the results thus far it is not possible to conclude that species-specific lifespan correlates with DNA-repair capacity. The lesson here is probably that we still lack comprehensive insight into the various ways an organism maintains its genome and how the various characteristics of the pathways to preserve the integrity of our genes impact on the process of aging (see below). In this respect, as we shall see, NER remains a very interesting system.
4.3.3.3 Mismatch repair (MMR) The third excision-repair pathway, MMR, is primarily devoted to the excision of mispaired nucleotides or short loops generated by insertions or deletions of nucleotides. These are typically errors made by the DNA-replicative machinery and MMR serves the purpose of correcting such errors. This makes MMR a postreplication repair system which complements an error-correction mechanism that is already part of many DNA polymerases. This built-in correction system is called proofreading and it is the primary line of defense against mistakes in newly synthesized DNA. When an incorrect base pair is recognized, DNA polymerase reverses its direction by 1 bp of DNA. The 3' → 5' exonuclease activity of the enzyme allows the incorrect base pair to be excised. Following base excision, the polymerase can re-insert the correct base and replication can continue. MMR, therefore,
GENOME MAINTENANCE
121
is an additional safeguard against errors during information transfer. The process is distinct from the two other excision-repair pathways in the sense that it relies on a way of distinguishing the newly replicated from the parental DNA strand. The mechanism by which MMR in eukaryotes is directed to the newly synthesized strand is as yet unknown. In E. coli, which has a very similar system to repair replication errors, the strand harboring the mispaired base is distinguished from the parental strand by adenine methylation on the latter. However, methylation in eukaryotes is different from prokaryotes and it is possible that part of the increased complexity of MMR in eukaryotes is due to the need for alternative mechanisms of strand discrimination. MMR in eukaryotes involves homologs of the E. coli MutS and MutL proteins, but is much more complex. A dimer of the MutS homologous MSH2 and MSH6 proteins (MutS) recognizes mismatches and single-base loops, whereas the MSH2–MSH3 dimer (MutS ) recognizes insertion and deletion loops. Then, the MutL-like protein complexes MLH1–PMS2 and MLH1–PMS1 bind to the MSH2–MSH6 and MSH2–MSH3 MutS homologs, respectively, after which the complexes can migrate in either direction. It is not only strand discrimination that remains unclear in eukaryotes; the excision step does as well. Evidence has been obtained, using a cell-free system, that the PCNA/RFC clamp and clamp-loader pair interact with MSH and MLH proteins to direct excision of the mismatch on the nascent strand through exonuclease327. The details of this process are still unknown in mammalian cells in vivo. Errors in DNA replication are not the only substrate for MMR. MMR plays a role in preventing strand exchange between divergent DNA sequences during HR, and in this sense functions as a monitor of meiotic and mitotic recombination. MMR acts as a barrier against the exchange of DNA molecules from unrelated organisms by aborting mismatched heteroduplexes. MMR is also capable of recognizing DNA damage other than mismatches. An example is its action in stabilizing hairpin structures that arise during BER of SSBs in repetitive stretches of microsatellite triplet repeats328. Such action may allow repeat expansion in postmitotic cells, such as in the brain of patients with Huntington’s disease (see Chapter 3). Lack of MSH2 interrupts the progressive repeat expansion that has been demonstrated in the mouse model for this disease329. As already briefly discussed in the section on DNA-damage signaling, MMR appears to be active also in DNA-damage signaling, including the signaling of an apoptotic response. Finally, Donald MacPhee (Melbourne, Australia) has suggested that MMR also acts in nondividing cells in the repair of mismatches generated during error-prone repair or as a consequence of mitotic recombination330. Repair of such mismatches by MMR in a non-dividing cell would result in mutations some 50% of the time because there is no way to distinguish the correct from the wrong base. Whereas NER deficiencies will always be associated with XP, MMR genetic defects are responsible for hereditary non-polyposis colorectal cancer (HNPCC), a human autosomal dominant disorder characterized by a strong predisposition to develop tumors, most
122
GENOME MAINTENANCE
notably tumors of the colon331. Most HNPCC cases are caused by mutations in MLH1 or MSH2. Cells from tumors of these patients display instability of the small repeat-element clusters called microsatellites (see Chapter 3). This is due to the tendency of the replication machinery to slip when encountering stretches of short repeat elements, resulting in the loss or addition of one or more repeat units. Such slippage replication errors result in the short insertion or deletion loops, which can only be repaired by MMR. We know that MMR defects are associated with microsatellite instability because we can readily detect such events from analyzing the tumor DNA. The single germ-line mutation in an MMR gene in an individual with HNPCC does not result in MMR deficiency. It is a second hit in the remaining functional allele in a tumor that eliminates MMR and results in such high levels of alteration in the notoriously unstable microsatellites that it can be detected directly. However, this microsatellite instability reflects a much broader genomic instability, known as a mutator phenotype, which helps the tumor cell to acquire new attributes for unhindered growth (see also Chapter 6). This explains the predisposition to cancer of MMR genetic defects in HNPCC. Indeed, genomic instability and predisposition to cancer are hallmarks of MMR defects, more so than in the case of NER defects. This is best illustrated by the MMR-deficient mouse models that have been generated. MMR-defective mice, as a consequence of the inactivation of any of the MutS or MutL homologs, develop both lymphomas, a common spontaneous tumor in normal mice at advanced age, and intestinal tumors. These mice usually die of cancer as early as 6 months of age332. To study mutations other than microsatellite instability, MMR-defective mice were crossed with other mouse strains harboring bacterial reporter genes that can be recovered from the mouse genome and tested in E. coli for mutations (see Chapter 6 for a detailed explanation of such mouse models). Such reporter loci are more similar to gene sequences than microsatellites and therefore a better marker for overall genome instability. Using such hybrid mouse models it was found that this high susceptibility to cancer of MMRdefective mice is associated with dramatically increased levels of genomic instability. For example, in the absence of any mutagenic treatment, mice lacking PMS2 showed a 100fold elevation in mutation frequencies in all tissues examined, as compared with the wildtype or heterozygous animals333. Most of these mutations were deletions and insertions of one nucleotide within mononucleotide repeat sequences, which is consistent with the role of MMR in the excision of mismatches caused by slippage during replication. As we have seen in the case of NER, it is often informative to model human diseasecausing mutations in the mouse, rather than generating complete knockouts. Indeed, in many cases these disease-causing mutations do not completely eliminate gene function, but only affect part of it. MMR mouse lines have been generated that harbor specific mutations causing HNPCC in humans. Analysis of cells from such knockin mice revealed that whereas MMR knockouts lacked both mismatch repair and the apoptosis response, an MSH2-specific mutation inactivated only the former.Whereas the mutant MSH2–MSH6
GENOME MAINTENANCE
123
complex retained normal mismatch recognition and, hence, can presumably still signal apoptosis, it lost its capacity to initiate the next stage in MMR. To retain the apoptotic response is obviously advantageous because mutant cells that fail to be repaired by MMR can be eliminated. This is in keeping with an observed significantly longer lifespan of these mice compared with the complete MMR-knockout mice. Nevertheless, eventually they also succumbed to cancer (about 6 months later) and their cells also displayed increased mutation rates334. Interestingly, this separation of the repair function from the DNA-damage signaling function of MMR would rule out the futile repair hypothesis described in section 4.2.6. MMR-deficient mice show increased predisposition to cancer but not premature aging. In this respect, it is of interest to compare the MMR-deficient mouse models with the previously discussed NER models. Whereas a deficiency in NER can also predispose to cancer (as, for example, in XP), several NER mutants are not cancer-prone. Some, including mice with a mutational defect in the XPD gene, which in humans causes trichothiodistrophy, are even resistant to cancer335. There are two possible explanations for this difference. First, as we have seen, MMR-defective mice lack an apoptotic response, an important barrier against cancer. In NER-deficient mice this response is not only retained, but likely plays a major role in reducing cancer in some of these mutants and promoting some of the degenerative aspects of aging (Chapter 5). Second, as we have seen, MMR defects can result in very high levels of genomic instability, much higher than what has been observed in most NER-deficient mice. It is likely that this increased mutagenesis, which mainly involves small mutations, is the cause of the greatly elevated cancer risk. This is because it increases the chance of mutations in genes that are critical for cancer development or progression. However, point mutations may contribute less to the non-cancer-related, degenerative aspects of aging. As we will see later in the book, it is possible that this non-cancer component of the aging phenotype is caused by other types of random genome alteration, such as large genome rearrangements. Such events are introduced into the genome as a consequence of erroneous repair of DNA double-strand lesions. It is to the repair of this type of DNA damage that I will now turn.
4.3.4 REPAIR OF DNA DOUBLE-STRAND LESIONS DNA double-strand lesions are forms of DNA damage that affect both strands simultaneously at opposite sites that are sufficiently close together to prevent the use of excision repair. The model lesion for the types of repair that are usually employed to overcome double-strand lesions is the DNA double-strand break DSB; that is, when two or more breaks are formed in opposite strands of DNA within about 10–20 bp of each other. DSBs are induced by ioniz-ing radiation or ROS, and can also arise as a consequence of replication-fork collapse—when encountering DNA damage or difficult-to-resolve secondary
124
GENOME MAINTENANCE
structures. The best example of the latter is the passing of a replication fork through a SSB, which will then be converted to a DSB on one of the sister chromatids. DSBs are also generated deliberately, as part of the process of V(D)J recombination in B and T lymphocytes, to generate antigen-binding diversity in the immunoglobulin and T-cell receptor proteins. In this process DSBs are generated at specific sites by a nuclease composed of the RAG1 and RAG2 proteins. After these controlled, site-specific rearrangement events, the repair of the resulting DSBs employs the same non-homologous end-joining (NHEJ) process used by the cell to repair randomly induced, so-called illegitimate, DSBs (see below). DSBs and double-strand lesions in general are highly toxic and their correct repair is critical for the cell to survive. Incorrectly repaired or persistent DSBs can cause chromosomal aberrations, such as deletions, insertions, and translocations, the spontaneous accumulation of which may play a causal role in age-related cellular degeneration and death. Most DSBs are likely to be repaired by either HR or NHEJ. Of these two processes, HR is generally thought to be error-free since it employs a homologous DNA molecule, usually a sister chromatid, as template to overcome the break. As will be described in more detail below, NHEJ merely joins the ends of breaks without the option of distinguishing one end from another when a cell suffers from more than one DSB at the same time. As for all DNA-repair pathways, there are strong interactions between DNA-damage sensors and the repair of DSBs, in this case especially ATM. The DSB signaling cascade, involving ATM, ATR, DNA-PK, and other proteins, leading to the activation of various downstream substrates, such as p53, has already been discussed. Although the details regarding the coordination between DNA-damage sensors and the various DNA-repair executive branches are not clear yet, components of DNA-damage signaling are known to be involved in repair per se. The role of ATM and especially DNA-PK in activating NHEJ has already been mentioned. In addition, ATM phosphorylates active participants in HR, including BRCA1. The DNA-damage checkpoints are also thought to be responsible for the p53-dependent elevation of the levels of deoxyribonucleotides in the nucleus to facilitate the DNA synthesis steps of DSB repair336. Most importantly, DNA-damage sensors are thought to play a major role in facilitating access of the repair enzymes to the sites of damage by reorganizing chromatin structure. We have seen that this is true for the excisionrepair pathways and it applies equally well to DSB repair. As already discussed, chromatin remodeling as a prelude to DSB repair is triggered in mammalian cells by phosphorylation of H2AX by ATM or DNA-PK. Finally, if all else fails, the damage signaling pathways may lead a cell into apoptosis or cellular senescence. Since DSBs are so toxic, it is likely that they contribute significantly to both spontaneous apoptosis events and the accumulation of senescent cells over a lifetime in a given mammalian tissue. As will be discussed in Chapters 5 and 7, apoptosis and cellular senescence are two of the three major cellular endpoints contributing to aging, neoplastic transformation being the third. Unfortunately, we do not yet know how a cell decides to enter apoptosis or cellular senescence. It is possible that, similar to the decision to abandon repair, it is the amount of damage that dictates whether a cell should die or enter a state of permanent replication arrest.
GENOME MAINTENANCE
125
Below I will briefly discuss the two major pathways for repairing DSBs and the physiological consequences associated with their action or lack of action. For more detailed discussions of the repair of double-strand lesions in general, including the highly toxic interstrand cross-links, the reader is referred to several excellent recent reviews337–340.
4.3.4.1 Repair by homologous recombination (HR) Repair by recombination is probably the most ancient form of repair in the living world. It has been proposed that sexual reproduction, which is essentially HR contributing to the generation of genetic diversity, arose very early in evolution as a way of overcoming damage in the genome165. A model for an early form of recombinational repair has been presented by Michael Cox (Madison, MI, USA)164, which is essentially a form of postreplication repair, applicable to either an RNA or DNA genome, that requires an homologous double-stranded nucleic acid from elsewhere. Most of our knowledge of HR as a mechanism to overcome DSBs stems from work with bacteria and yeast. This has resulted in the elegant DSB-repair model for meiotic recombination by Szostak et al.341. The pathway is obviously well conserved and critically important in mammals. The mechanism of the repair of DSBs by HR in mammalian cells is essentially similar to the Szostak model or the ancient form of HR proposed by Cox. Before strand invasion, the first major step in HR, the ends of the DSB undergo resection in the 5' to 3' direction by the 5' to 3' exonuclease activity of the MRN complex (see section on DNA-damage sensors). This results in 3' single-stranded tails, the targets for a complex of proteins originally discovered in yeast and known as the RAD52 group. Mammalian homologs of all the factors in this group have been described. From the molecules in this group, it is the RAD51 protein that acts as the mammalian counterpart of E. coli RecA by forming a nucleoprotein filament that coats the single strands and invades the exchange partner of the damaged DNA. Once identified, the homologous sequence is subsequently used as a template for DNA synthesis to extend the single-strand DNA tails. The strands are sealed to the correct parental strand by DNA ligase. The resulting Holliday junctions are resolved by resolvases to yield two intact DNA molecules. This leads to crossover and non-crossover products, depending on the mechanism of resolution. As we will see later, the protein helicase encoded by the BLM gene, which is defective in the human disease Bloom syndrome, plays an important role in the resolution process. There is still considerable debate about the details of HR, which may not always be the same and could also differ from species to species. The entire process, which involves many different actors, including the above-mentioned BLM helicase and the breast cancer-susceptibility proteins BRCA1 and BRCA2, is concentrated in nuclear foci. These foci can be visualized under the microscope after fluorescent staining with antibodies against one or more of the participants. In principle, HR is error-free, since it uses a clean copy of the damaged DNA molecule as a template. The ideal template in this respect is a sister chromatid, which is why HR is
126
GENOME MAINTENANCE
thought to be particularly active in repairing DSBs that are induced during the S/G2 phase of the cell cycle. However, HR could also use the homologous chromosome as its exchange partner during the G1 phase of the cell cycle, when a sister chromatid is not present. This requires searching around in the nucleus for the homolog, a cumbersome process which may explain why G1 cells prefer to repair DSBs through NHEJ (see below). Another reason to avoid the homologous chromosome as template for HR is the possibility that mutations arise as a consequence of crossing-over. In the absence of crossingover the repaired segment will simply acquire the donor sequence, a form of nonreciprocal transfer called gene conversion. With crossing-over, however, large parts of a chromosome are exchanged with a substantially increased risk of loss of heterozygosity (LOH); that is, the loss of the single functional allele when the other is already inactive due to a preexisting mutation. This is a major reason to avoid crossing-over when resolving Holliday junctions. Candidate proteins for resolving Holliday junctions in mammalian cells (and possibly also in yeast) without crossover products include the protein encoded by the BLM gene, a member of the RecQ family (see below). In spite of the penalty, mitotic recombinations do occur as demonstrated342, indicating that HR does occasionally use the homologous chromosome as template, thereby generating crossover products. However, they would probably occur much more frequently were it not for its apparent suppression by MMR components on the basis of sequence differences between the potential exchange partners. This effect of increased genetic distance has been demonstrated by a comparison of progeny from crosses between strains of mice with significantly diverged nucleotide sequences; in progeny of such strains, mitotic recombination is hardly detectable343 (see also Chapter 6). The impact of mitotic recombination on spontaneous levels of genomic instability during aging of mammals is unclear. In budding yeast, however, a dramatic increase in LOH events during replicative aging of this organism has been reported151. This phenomenon is unrelated to the accumulation of extrachromosomal rDNA circles in yeast mother cells (see Chapter 2), which appeared to cause replicative aging in at least some strains of yeast and may also arise as a consequence of illegitimate recombination events. The possible relationship of these two forms of genomic instability in yeast to the aging process will be discussed extensively in Chapter 7. Another disadvantage of HR repair in G1 involves the possibility of a DSB occurring in any of the large number of repeat elements in the mammalian genome. In the absence of a conveniently lined-up sister chromatid, this could easily lead to the selection of a repeat family member on another chromosome as exchange partner. This would result in chromosomal translocations. When the exchange partner is a repeat family member on the same chromosome this would result in deletions, a potential mechanism for the appearance of extrachromosomal circular DNA in aging yeast mother cells. Whereas in mammals there is no evidence that extrachromosomal circles accumulate with age as in yeast, extrachromosomal DNA is common and possibly also a consequence of erroneous
GENOME MAINTENANCE
127
recombinational repair152 (see also Chapter 6). To limit such instability, HR is restricted to late S and G2, but it is nevertheless possible that a significant portion of the chromosomal aberrations as they are observed in human and animal cells are a result of such illegitimate recombinations. Finally, in an error-prone variant of HR DSBs are repaired in a pathway that does not use the sequence information from a sister chromatid or homologous chromosome. This mechanism is called single-strand annealing. Like HR, single-strand annealing requires the presence of DNA sequence homology, in this case in the form of complementary sequences on both sites of the break. The process begins very similar to HR by resecting the 5' ends of the break. This resection exposes complementary regions within the 3' overhangs, usually repeat sequences. In mammalian cells such repeats are of course frequently present. Annealing takes place at these repeats with the loss of the flap-overhangs (by a FEN1-like nuclease). Loss of the sequence between the repeats is the inevitable result of single-strand annealing, which is therefore always mutagenic. Apart from DSBs, HR is also involved in repairing the highly toxic interstrand crosslinks339. Interstrand cross-links are induced by chemotherapeutic agents, such as mitomycin C, and cisplatin, but also by natural compounds in plants as well as endogenous agents formed during lipid peroxidation. To effectively remove interstrand cross-links, different repair pathways have to act in concert. Repair of interstrand cross-links in E. coli has been well characterized and is based on incisions by the NER enzymes UvrABC, followed by HR. Alternatively, a combination of NER and translesion synthesis is employed. As expected, in yeast and mammalian cells the situation is somewhat similar, but more complicated. In mammalian cells it is unknown how the interstrand cross-links are recognized. Also the precise role of the NER enzymes in mammalian cells is unclear. However, Chinese hamster ovary cells with defects in the endonuclease ERCC1–XPF are exquisitely sensitive for agents that induce interstrand cross-links. Moreover, ERCC1knockout mice have a much more severe phenotype than other NER mutant mice. They are not only UV-sensitive, but display premature aging symptoms and live only for a few weeks (for a more detailed discussion of this mouse model, see Chapter 5). Like the ERCC1–XPF mutant Chinese hamster ovary cells, cells from these mice are very sensitive to mitomycin C and other agents that induce interstrand cross-links. There is no evidence that XPG, the other endonuclease active in NER, plays a role in interstrand cross-link repair. The role of ERCC1–XPF in repair of interstrand cross-links is unclear and it may be based on its function in removing the non-homologous DNA tails during HR. Defects in interstrand-cross-link repair have been implicated as the cause of Fanconi anemia, a rare, autosomal recessive disorder, with developmental abnormalities, progressive bone-marrow failure, and a greatly increased risk of cancer344. The disorder is characterized and diagnosed by the sensitivity of the patient’s cells to agents inducing interstrand crosslinks. Fanconi anemia is a genetically heterogeneous disorder with as many as 12 complementation groups identified, corresponding to defects in distinct genes involved in the
128
GENOME MAINTENANCE
genome-maintenance pathway represented by this disease. Some protein products of the identified genes have been shown to functionally or physically interact with BRCA1, RAD51, and the MRN complex. Indeed, one protein, FANCD1, was identified as BRCA2, a human breast cancer-susceptibility gene. Like BRCA1, this gene has been implicated directly in the process of HR repair (see above). The function of most of the Fanconi anemia proteins is still unclear, as is the exact nature of the condition, which is also characterized by symptoms of oxidative stress.
4.3.4.2 Non-homologous end-joining (NHEJ) In yeast, HR is the preferential pathway to repair DSBs, although NHEJ also occurs. This is different in mammals, in which NHEJ is extensively employed. This difference in preference is likely to be due to the greatly increased time mammalian cells in vivo spend in G1 or G0. Indeed, NHEJ seems to be especially important in differentiated cells; early in development, HR is probably more important. In mammals, NHEJ is also used to repair the RAG-generated DSBs in the process of V(D)J recombination. This can only occur by NHEJ, as indicated by the observation that inactivation of each of the six NHEJ genes in humans or mice prevents B- and T-cell maturation, resulting in severe combined immunodeficiency (SCID)345. The main player in NHEJ is the Ku heterodimer with its 70- and 80-kDa subunits, known as Ku70 and Ku80, respectively. Ku80 is sometimes also referred to as Ku86. As we have seen, Ku is a part of DNA-PK, which consists of the Ku heterodimer and DNA-PKCS, the catalytic component with kinase activity upon binding to DNA ends. Ku, which is basically a ring, binds to the DSB by threading its ends. In this way it holds them together, probably protects them against nucleolytic attack, and also serves as the recruiting platform for the repair proteins, including DNA-PKCS. The role of the latter in NHEJ is not exactly clear. We have already seen that this giant protein kinase of the PIKK family, like ATM and ATR, plays a role in DNA-damage sensing and signaling, especially in signaling the apoptotic response. It also plays a role in telomere maintenance and it is activated by bacterial DNA as part of the innate immune response. In NHEJ its function must be specific for vertebrates since they are the only organisms in which it has been identified. Two molecules of DNA-PKCS on each side of the break may assist Ku to hold the DNA ends together. The structures of these ends, however, are not amenable to a simple ligation (as is also usually the case with SSBs; see above under BER). They may have singlestranded overhangs and damaged bases or sugar moieties that require processing. It is likely that the protein called Artemis plays an important role in resolving the various DNA end modifications346. Artemis binds to DNA-PKCS and is phosphorylated. This stimulates and extends the nuclease activity of Artemis, so that the protein becomes capable of cutting away protruding single-stranded regions at DNA ends and creating
GENOME MAINTENANCE
129
double-stranded structures that are good ligase substrates. Other factors, such as the Werner syndrome protein (WRN) and the previously discussed MRN complex may play a role. The polymerases needed to fill the gap are those belonging to the X family. These enzymes, such as pol , pol , or the terminal deoxytransferase (TdT), are apparently utilized depending on the template. DNA ligation is subsequently carried out by ligase IV in a complex with XRCC1. NHEJ is active throughout the cell cycle in all vertebrate tissues and only competes with HR in the S and G2 phases. In contrast to HR, NHEJ is error prone. Part of the necessary processing of DNA ends before they can be religated is resection. We have seen this happening in HR, which is not a problem since it relies on its identical sister chromatid to retrieve the lost information. In NHEJ, however, such resection leads to the permanent loss of between 1 and 20 nucleotides each time a DSB is repaired. Realizing the potential impact of such a continual loss of small amounts of genomic sequence over the lifetime of an organism, Michael Lieber (Los Angeles, CA, USA) has made an attempt to estimate its consequences in terms of the loss of gene activity of an average cell347. Based on his estimate of 5–10% of all cells harboring a DSB at all times and a repair time of 24 h, he calculated that at the end of a human life each cell would have 2300 imprecise repair sites distributed throughout the genome. In a genome with 50 000 genes (we now know that this is probably less) and only 5% of the genome functional (a very low estimate; Chapter 3), he reached the not inconsiderable number of 430 genes being adversely affected. A second potential source of errors associated with NHEJ is its lack of a mechanism to distinguish DNA ends corresponding to different DSBs arising in the same cell at the same time. It is likely that NHEJ is responsible for random integration of transforming DNA in transgenesis via sporadic DSBs in the chromosomes. It has been demonstrated that by increasing the number of DSBs in a cell from one to two the frequency of translocations is increased by at least 2000-fold348. The situation will be worse with more DSBs in a single cell. Using H2AX as a marker for multiple DSBs induced by a linear track of particles, Aten et al.349 observed congregation of these breaks into small clusters in G1-phase cells. This supports the notion that distant DSBs can be juxtaposed, which would greatly increase the generation of translocations. Whereas we may assume that Ku would be quick enough to capture the ends of each DSB before they can float apart, the promiscuity of NHEJ could lead to chromosomal translocations, especially when multiple DSBs are juxtaposed. Interestingly, whereas one might think that inactivation of NHEJ would prevent such translocations (DNA ends cannot ligate to one another by themselves) its absence in the mouse has been associated with increased chromosomal aberrations350. It is possible that in such a situation HR functions as a back-up, inaccurate in G1 as described above, or single-strand annealing of complementary single strands from different chromosomes. Finally, non-classic forms of end-joining could be active that have thus far been elusive.
130
GENOME MAINTENANCE
4.3.4.3 Effect of DSB repair defects in animals As for all genome-maintenance pathways, mouse models have been made in which HR or NHEJ genes were inactivated. The results indicate that deficiencies in critical HR genes are usually fatal, but those in genes involved in NHEJ generally not. For example, inactivation of key genes involved in the crucial strand-invasion step of HR, for example, RAD51 BRCA1, and BRCA2, is embryonically lethal. This lethality is understandable in view of the role HR plays in repairing the DSBs that must arise when the replication fork meets a SSB. By contrast, most gene deficiencies in core NHEJ genes, for example Ku70, Ku80, and DNA-PKCS, result in viable mice. This lack of lethality must be due to the relative scarcity of non-dividing cells during development. Indeed, the animals do suffer from SCID (due to the defect in V(D)J recombination) and they show multiple symptoms of accelerated aging (see Chapter 5).
4.3.5 ANCILLARY SYSTEMS In addition to the main DNA-damage checkpoint and repair pathways discussed above, a number of systems have been identified for resolving particular problems related to DNA damage and genome instability. These systems often closely interact with the DNA-repair systems per se; that is, those that act to remove lesions, such as the excision-repair systems. Here I will discuss three of such ancillary systems: telomerase, RecQ helicases, including WRN, and DNA topoisomerases. Inactivating mutations in genes participating in these systems in the mouse result in phenotypes resembling accelerated aging, similar to the situation for NHEJ and NER.
4.3.5.1 Telomere maintenance As discussed in Chapter 3, mammalian telomeres are composed of TTAGGG repeat arrays bound by a complex of proteins and organized in a so-called T loop (Fig. 4.7a). This prevents attack by DNA-repair enzymes, which would otherwise recognize these chromosome ends as DSBs. In contrast to bacteria, which have a circular genome, eukaryotes are unable to complete lagging-strand synthesis during replication (see Chapter 3 for a detailed description). As a consequence, 50–200 bp of telomeric repeats are lost with each cell division. The telomerase reverse transcriptase (TERT) maintains telomere length by copying a short template sequence within its intrinsic RNA moiety (TERC), running 5' to 3' towards the end of the chromosome. It extends the 3' tail, after which normal lagging-strand synthesis generates the opposite strand (Fig. 4.7b). This process is controlled by regulator proteins, some of which (TRF1, TIN2) inhibit telomerasemediated lengthening to prevent unrestrained elongation of telomeres.
GENOME MAINTENANCE
(a)
T loop
G-rich tail
(b)
131
TRF1-associated proteins 3
TRF2
5 3
3 5 Telomerase
5 3
3 5 DNA synthesis 3
5 3
5
Fig. 4.7 (a) The telomeric ends of chromosomes are organized in the form of T loops. (b) Action of telomerase as it corrects the effects of the end-replication problem.
Both telomeres and telomerase are important genome-maintenance systems. The T-loop structure prevents NHEJ from fusing different chromosomes together and telomerase prevents telomeres from becoming so short that they can no longer form T loops. Telomeres that shorten to the point that they become dysfunctional cause replicative senescence by activating p53. Inactivation of the p53 pathway allows cells to bypass replicative senescence and proliferate until they reach a stage called crisis, which is characterized by widespread chromosomal instability (with its characteristic telomeric end-to-end fusions), apoptosis, mitotic catastrophe, and senescence, with the rare emergence of immortal cell clones all of which have acquired a mechanism to re-stabilize their telomeres. The mechanism of such genomic instability is not entirely clear but NHEJ is the most likely culprit351. Inactivation of the p53 pathway is enough to bypass senescence due to telomere dysfunction. Normal human cells do not express telomerase at all, except embryonic tissues, germline tissues (testes and ovaries), and progenitor cells in tissues with high cell turnover, such as bone marrow and intestinal crypts. As mentioned above, in the majority of human
132
GENOME MAINTENANCE
cancer cells telomerase activity is easily detectable. Such cells exhibit stable telomere lengths upon extended propagation in culture and do not undergo replicative senescence. In mouse cells, the situation is different in the sense that most laboratory strains have very long telomeres and TERT, the gene encoding the telomerase reverse transcriptase component, is widely expressed at low levels in adult tissues, with greatest abundance during embryogenesis and in adult thymus and intestine352. The continued availability of telomerase in mice could be a major factor determining its short lifespan as compared to humans. Cancer, a major aging-related disease, is kept at bay much longer in humans than in mice (in absolute time), possibly because of the strict regulation of telomerase as a major gatekeeper to prevent immortalization. The difference in telomere length and telomerase regulation also explains, at least in part, why senescence in mouse cells is fundamentally different from senescence in human cells353. In humans, even a partial loss of telomerase function already causes serious disease. This is illustrated by the human inherited disorder dyskeratosis congenita, an autosomal dominant form of which is caused by mutations in one of the two copies of the gene that encodes TERC, the RNA part of telomerase. These patients suffer from abnormalities in the production of cells in the blood and in the gut, and have poor wound healing, early baldness, hypogonadism, and lung fibrosis. Not surprisingly, they generally die early. From these symptoms it can be derived that the cells affected are progenitor cells in tissues with a high cell turnover, which is very similar to the situation in Terc-deficient mice. In such animals symptoms manifest only after several generations because telomeres of most laboratory strains of mice are so much longer than those of humans. Interestingly, whereas the short telomeres in Terc-deficient mice are associated with increased cancer incidence as a consequence of the genomic instability, some tissues and cell types seem to suffer less from cancer as compared with normal control mice. For example, late-generation Terc-null mice are resistant to skin tumorigenesis354. Reduced cancer has also been observed in mice deficient in both telomerase and p16 (INK4a)355. This may reflect slow tumor progression in these tissues due to high levels of tumor cell death as a consequence of their short telomeres, which would be in keeping with the need of most tumors to stabilize telomeres, which is most easily accomplished by induction of telomerase (see above). It is important to notice that these mice have intact DNA-damage responses. Hence, whereas the increased genomic instability accelerates tumor formation, the constitutive activation of DNA-damage response factors, such as p53, will greatly impair their progression by inhibiting cellular proliferation. Impairment of cellular proliferation through short telomeres will also adversely affect stem-cell reservoirs, which may explain the premature appearance of multiple aging-related phenotypes in Terc-null mice. The antagonism between cancer and aging and its possible implications for the causes of normal aging will be discussed further in Chapters 5 and 7. Clearly, in tissues with high cell turnover telomerase is critically important. However, it is not the only system that can extend and maintain telomeres.As first discovered in yeast356, est1 mutants (lacking telomerase) can adapt to telomere erosion by two mechanisms
GENOME MAINTENANCE
133
termed type I and type II survival. Type I survivors require RAD51, a sure sign that HR is involved357. Indeed, in mammals, the so-called alternative lengthening of telomere (ALT) pathway was first identified in replicatively immortal, tumor-derived cell lines that do not express telomerase358. Dividing tumor cells cannot do without a mechanism to maintain their telomeres and generally manage to switch telomerase on. If telomerase expression cannot be restored, the ALT pathway is an alternative. Its activity can be recognized by the accumulation of so-called ALT-PML bodies, a variant of the PML bodies mentioned in Chapter 3, in this case consisting of DNA-repair proteins co-localized at telomeres. Interestingly, HR may play a role in normal telomere maintenance as well. This is suggested by the presence of RAD51D, one of the five RAD51 paralogs, at telomeres of normal mouse cells359. Its absence was shown to cause chromosomal aberrations, including typical end-to-end fusions. Such chromosomal instability was also observed in telomerase-negative immortalized human cells in which RAD51D was suppressed by siRNA359. Many different DNA-repair proteins have been localized at telomeres, including Ku, DNA-PKCS, BRCA1, BRCA2, and the MRN complex, as well as the RecQ proteins WRN and BLM (see below). It is possible that these repair factors play a role in maintaining telomeres, perhaps in association with telomere-specific proteins. This is somewhat paradoxical since many of these proteins are supposed to repair broken DNA molecules, presumably including the ends of uncapped chromosomes360. In yeast, proteins of the NHEJ system may be intimately involved in maintaining telomere length, as indicated by the shortened telomeres in cells lacking either of the two Ku proteins. Also in mammals there is evidence for DNA-repair proteins playing a role in normal telomere maintenance, but the details of this involvement are still incompletely understood. Alternatively, telomeres could merely be a storage site for DNA-repair proteins, similar to nucleoli and PML bodies. Whereas we now know that repair proteins are located at telomeres and may even play a role in telomere maintenance, the recognized telomere cap proteins, such as the telomere repeat factors TRF1 and TRF2, are clearly the most important in this respect. As already mentioned, TRF1, together with its interacting partners TRF2, TIN2, TANK 1, and TANK 2, are regulators of telomere length361. Some of these factors negatively control telomere length by promoting a telomeric architecture that limits the ability of telomerase to access telomeres. For example, dominant negative forms of TIN2 or TRF1 can extend telomere length362. The importance of such regulation is illustrated by lack of viability of TRF1-deficient mice363. Interestingly, this embryonic lethality could not be rescued by telomerase, suggesting that it is not due to uncapping or attrition of telomeres. TRF2 has mainly been implicated in the formation of the T-loop structure and in preventing repair enzymes from recognizing telomeres as DNA DSBs. After TRF2 deletion in mouse cells, the telomeres lose the 3' overhang and are processed by the NHEJ pathway. Overexpression of TRF2 in the mouse skin was found to resemble the NER disorder XP, with keratinocytes from these animals hypersensitive to UV and DNA cross-linking agents364. Skin cells of these mice were also found to have short telomeres and loss of the
134
GENOME MAINTENANCE
G-strand overhang. It was suggested that the increased TRF2 raises the activity of XPF (with which it is known to interact, as with many other repair proteins) at telomeres, possibly at the cost of XPF activity elsewhere in the genome, which may explain the increased UV sensitivity. Evidence is now emerging that whereas general DNA-repair proteins are important for telomere maintenance, telomere-specific proteins may be important for genome maintenance in general. Interestingly, TRF2 was recently found to quickly and transiently localize at DSBs induced by a laser microbeam at all possible locations in the genome365. This suggests a general role for TRF2 in genome maintenance, possibly in preventing premature action of DNA-repair enzymes until the correct repair complex is assembled. How important is telomere maintenance in the context of maintaining the genome overall? As I will discuss in more detail in Chapter 6, telomere length measurements, mainly in white blood cells, have revealed that telomeres shorten during human aging. However, age-adjusted telomere length is highly variable, possibly because of heritable factors and/or disease. This considerable individual heterogeneity and the overlap in telomere lengths between young and elderly individuals renders any correlation weak and the significance unclear. Nevertheless, a relationship between enhanced telomere attrition, reduced cancer, and a more pronounced manifestation of non-cancer, degenerative symptoms of aging can be rationalized. Telomere shortening protects us from cancer by impairing tumor cell proliferation, through cell-cycle arrest or by activating replicative senescence and apoptosis. However, this would also result in a loss of functional cells, which in turn can be expected to lead to age-related organ dysfunction. Adverse effects of telomere shortening may be especially prominent in stem-cell reservoirs. Indeed, it has been demonstrated that telomere shortening (in Terc-deficient mice) inhibited mobilization of epidermal stem cells out of their niche in hair follicles, whereas telomerase overexpression promoted stem-cell mobilization366. The latter is apparently not related to telomere lengthening but a reflection of another TERT functional pathway367. On the other hand, overexpression of TERT in mouse hematopoietic stem cells had no effect on the transplantation capacity of these cells; irrespective of telomerase activity, such cells could be transplanted for no more than four generations368. Of note, the situation could be very different in humans due to a difference in telomere biology between the two species. Nevertheless, it is possible that whereas telomere attrition is the main barrier of replicative capacity of human cells in culture, it may not necessarily exert a major influence on cells in vivo.
4.3.5.2 RecQ helicases and DNA topoisomerases Members of the RecQ helicase family occur in organisms varying from prokaryotes to mammals and are named after the RecQ protein originally identified in E. coli. In mammals there are five RecQ helicases, defects in three of which have been associated with genomic
GENOME MAINTENANCE
135
instability, cancer, and premature aging in humans in the form of the heritable, segmental progeroid disorders Bloom syndrome, Werner syndrome, and Rothmund–Thomson syndrome369. The RecQ homologs share a central helicase domain, with only the Werner syndrome protein WRN harboring an exonuclease activity as well. It is also Werner syndrome, and to a lesser extent Rothmund–Thomson syndrome, in which premature aging is especially prominent; Bloom syndrome is mainly characterized by excessive tumor formation (see Chapter 5 for a more detailed discussion of segmental progeroid syndromes). Most of what we know about RecQ helicases in mammals is derived from the properties of the BLM and WRN proteins. RecQ homologs are able to catalyze unwinding of many different DNA substrates, such as forked DNA, Holliday junctions, and triple and tetraplex DNA (see Chapter 3). Whereas all the probably multiple roles of RecQ homologs in genome maintenance are not yet known, they are generally thought to be ancillary factors in replication, recombination, or repair by resolving secondary structures. The best-characterized RecQ protein in this respect is BLM. As established by Ian Hickson (Oxford, UK) and collaborators, one function of BLM, in concert with topoisomerase III(TOPIII; see below) is to resolve recombination intermediates containing double Holliday junctions by a process they called double Holliday junction dissolution370. This function is apparently specific for BLM since WRN and other RecQ homologs cannot substitute for BLM in these dissolution reactions. Human Bloom-syndrome cells show a characteristically elevated level of spontaneous exchanges between sister chromatids and homologous chromosomes, which indicates increased HR. BLM’s resolution of recombination intermediates avoids crossover events during HR. It was proposed that this would protect organisms with large genomes, containing repetitive sequences, against mutagenic genome rearrangements as a consequence of exchange between non-sisters. The danger of such erroneous forms of HR in generating LOH events or other types of genomic mutation has already been mentioned in the section on HR. Since Bloom-syndrome cells are not hypersensitive to ionizing radiation, indicating that BLM is not essential for the repair of DSBs, it was proposed that crossover suppression through this mechanism was especially important in the repair of daughter strand gaps arising during replication of a damaged template. As discussed earlier, this postreplication mechanism for damage avoidance relies on HR. Although none of the other RecQ family members is able to substitute for BLM in suppressing crossing-over through Holliday-junction dissolution, they may all be involved in decreasing genomic instability during some form of DNA transaction. This would explain the increased genomic instability observed in cells from Bloom syndrome and Werner syndrome as well as the high cancer incidence. It should be noted that mutant cells from both Werner syndrome and Bloom syndrome are viable and the existence of the human syndromes shows that impairment of RecQ functions is not incompatible with life, but merely increases spontaneous levels of genomic instability and cancer. However, in mice complete BLM deficiency is embryonically lethal371,372 and it is likely
136
GENOME MAINTENANCE
that this is true also in most human cases. Indeed, Bloom syndrome is much rarer than expected from the number of heterozygotes. So it is likely that only some allelic combinations allow for pre- and postnatal survival. WRN-deficient mice show virtually no phenotypic differences from their wild-type counterparts, unless they also harbor null mutations in Terc, Atm, or p53 (see Chapter 5). Whereas BLM and WRN display a vast overlap in function in vitro, there are significant differences in vivo. Of the two syndromes, Bloom syndrome is the most severe, with death usually before the age of 30. Werner syndrome patients die about 20 years later, mostly of cancer and cardiovascular disease. Cancer in Werner syndrome develops much later than in Bloom syndrome and is restricted to tumors derived from mesenchymal cells. Cells from Werner syndrome do not show the greatly increased rate of sister chromatid exchange as Bloom-syndrome cells. There is evidence that the WRN protein is required for maintaining telomere function during replication. Cells lacking WRN have been demonstrated to suffer from a loss of telomeric sequences from a single sister chromatid, which was explained by a defect in lagging strand synthesis373. Telomere loss was dependent on the helicase function of WRN alone and could be counteracted by expression of telomerase. WRN has an exonuclease function that BLM lacks. It is possible that the exonuclease activity is needed to process telomeric DNA. Another difference between the two RecQ homologs is that WRN, but not BLM, plays a role in NHEJ. This can be derived from observations that WRN binds to the Ku70-Ku80 component of DNA-PK, which stimulates the exonuclease, but not the helicase activity of WRN374. There is evidence that WRN acts in optimizing the repair functions of NHEJ and HR375. Nevertheless, at least in mice, its role in these processes cannot be that important since WRN-null mice do not suffer from similar problems as other mouse models in which a NHEJ or HR core protein has been inactivated. RecQ proteins have many binding partners among DNA-repair and -replication proteins, as can be expected from their role in assisting in the seamless performance of these processes. A most frequently found interaction is with DNA TOPIII, which is a type I topoisomerase. Type I DNA topoisomerases can change the topological status of the DNA by introducing a DNA SSB and then transfer another DNA strand through this break. By doing this they change the degree of supercoiling of the DNA. Physical interaction between Sgs1, the yeast RecQ helicase, and BLM with TOPIII has been described for yeast and human cells, respectively. Deletion of the TOPIII homolog in yeast (TOP3) results in slow growth, hyper-recombination, genomic instability, impaired sporulation, and increased sensitivity to genotoxic agents376. Interestingly, this phenotype is suppressed by deletion of Sgs1, the yeast RecQ helicase, indicating that Sgs1 creates a deleterious topological substrate that needs to be resolved by TOP3377. In vertebrates, there are two isoforms of TOPIII, termed and . The human BLM protein functionally interacts with TOPIII. In mice, deletion of TOPIIIis embryonically lethal. By contrast, mice lacking TOPIII are
GENOME MAINTENANCE
137
viable and grow to maturity with no apparent defects. However, once adults these animals show a reduced lifespan and display lesions in multiple organs resembling premature aging (see Chapter 5). It is conceivable that this effect is due to impairment of a RecQ family member rather than a direct consequence of the lack of TOPIII . Thus far, there is no evidence that WRN interacts with TOPIII. However, it has been demonstrated that the WRN protein physically and functionally interacts with TOPI378.
4.3.6 DNA REPAIR AND CHROMATIN Nucleosomes are severe obstacles for processing of DNA lesions and must be disrupted to allow repair. This has been studied most intensively for NER, especially after UV damage. Since the chromatin remodeling events necessary for disruption and restoration of nucleosomal structure associated with DNA repair are likely to depend on the type of lesion and repair pathway involved, we are still far from a comprehensive insight into these processes and their impact on genome stability in aging organisms. As we have seen, chromatin structure contains epigenetic information essential for genome functioning, which is encoded in the specific patterns of DNA methylation and histone modification. Are there chromatin-repair systems and what are the potential consequences of incomplete or erroneous restoration of higher-order DNA structure? One important player that has thus far been identified is the chromatin assembly factor CAF-1379. This multimeric protein specifically deposits histones onto DNA that has been subject to DNA-repair synthesis. This is very similar to chromatin assembly following DNA replication, which also requires the CAF-1 complex (Chapter 3). As in replication, CAF-1 recruitment depends on its interaction with PCNA. It has been demonstrated that CAF-1 is recruited to sites of NER (or SSB repair), but not in cells that are deficient in repair, such as XP cells. Yeast deleted for the gene encoding CAF-1 is highly sensitive to double-strand DNA-damaging agents380. It has been suggested that CAF-1 can also act as a chaperone in mediating the correct re-assembly of the histones displaced during repair. Indeed, another candidate chromatin repair protein is the histone chaperone anti-silencing function 1 (ASF1). It is unclear as to how and where this protein acts, but it is known to enhance the nucleosome-depositioning activity of CAF-1. In yeast, asf1 mutants are sensitive to DNA damage, suggesting its involvement in maintaining chromatin structure after repair381. Whereas it is unknown whether aberrant or incomplete chromatin repair can lead to loss of genome integrity through epigenetic changes, evidence for both incomplete restoration of DNA-methylation patterns after repair and methylation changes with age has been obtained (Chapter 6). Methylation patterns are established during embryogenesis and remain largely unchanged in adult cells382. Genome-wide demethylation occurs before embryo implantation followed by de novo global remethylation (by DNMT3a and
138
GENOME MAINTENANCE
DNMT3b de novo DNA methyltransferases) after implantation, before organ development. This allows gene-specific methylation patterns to develop, which determine tissue-specific transcription repression, including repression of either the maternal or paternal allele of imprinted genes. Once established during early life, methylation patterns are replicated during mitosis by the enzyme DNMT1, a maintenance DNA methyltransferase. However, methylation patterns are not completely stable. Gradual hypomethylation of the mammalian genome occurs with age in most tissues as well as aberrant hypermethylation in the promoter regions of genes (see Chapter 6). Interestingly, it has been demonstrated that after UV irradiation of non-dividing cells in culture restoration of the methylation patterns after the resynthesis step of NER was slow and incomplete383. This could explain the observed gradual demethylation associated with the aging process. It is unknown whether DNMT1 is also involved in replicating methylation patterns during DNA repair, but this is likely since evidence for an early accumulation of this enzyme (but not the de novo DNA methyltransferases) at sites of DNA damage has been obtained384. The observed incremental methylation of CpG islands with age of human colon385 could be due to aberrant de novo methylation.
4.3.7 PRE- AND POST-GENOME MAINTENANCE Naturally, systems for the prevention of DNA damage or the mitigation of its adverse effects are critically important parts of genome maintenance. A detailed discussion of the full repertoire of such systems is beyond the scope of this book. Here I will only briefly discuss the main systems for preventing oxidative DNA damage as well as the physiological buffering of the phenotypic penetrance of deleterious mutations.
4.3.7.1 Antioxidant defense In view of their abundance as normal by-products of metabolism, ROS, such as singlet oxygen, superoxide, peroxyl radicals, hydroxyl radicals, and peroxynitrite, are considered as probably the main source of spontaneous DNA damage386. However, ROS are also used by many cell types in normal processes, for example, by macrophages as part of their ability to defend the body against intruding microorganisms. To prevent ROS from rising to excessive levels, cells are equipped with a variety of antioxidant defense systems. Such systems include the enzymes superoxide dismutase (SOD), catalase, or glutathione peroxidase and a variety of dietary antioxidants. Antioxidants are compounds that protect cells against the damaging effects of ROS. The main dietary antioxidants are vitamin E,
-carotene, and vitamin C. An imbalance between antioxidant defense and ROS results in oxidative stress, leading to cellular damage. Oxidative stress has been causally linked to many diseases, including cancer, atherosclerosis, ischemic injury, inflammation, and neurodegenerative diseases, such as Parkinson’s
GENOME MAINTENANCE
139
disease and Alzheimer’s disease. As we have seen, oxidative stress has been implicated as the main causal factor in aging and antioxidant defense is therefore considered critically important in longevity assurance. However, the results obtained after inactivating antioxidant defense systems, such as SOD or catalase, or increasing their activity through overexpression of these enzymes or by supplementing antioxidants to the diet, are conflicting. Full or partial inactivation of antioxidant defense systems does not generally result in premature aging, which is in contrast to the aforementioned mice with inactivated DNA-repair systems (see also Chapter 5). Overexpression of antioxidant defense genes, such as SOD and/ or other ROS-scavenging enzymes, has been reported to increase lifespan in fruit flies140, but in the mouse there are few obvious beneficial effects. Indeed, in my own laboratory we have generated mice overexpressing both SOD1 and catalase as part of large, almost 100kb bacterial artificial chromosomes, providing most if not all the normal regulatory cisacting control elements. Whereas the activity of these enzymes was upregulated about 2-fold in all tissues analyzed387, the first results of lifespan studies do not indicate that they live significantly longer (A. Richardson, personal communication). In humans, clinical trials of antioxidant supplementation have failed to show benefit with respect to disease outcome and sometimes adverse effects were observed. However, these results are controversial because there is evidence that plasma levels of antioxidants are correlated with decreased disease risk388. In mice, no effect of long-term dietary supplementation with antioxidants on the pathological outcome or on mean and maximum lifespan has been observed389.
4.3.7.2 Mitigating the effects of mutations Although in general DNA mutations are irreversible their effect can sometimes be buffered. An example of such physiological buffering is the action of molecular chaperones, such as the heat-shock protein Hsp90390,391. Hsp90 assists with the maturation of many key regulatory proteins. Like other chaperones, Hsp90 recognizes and transiently binds hydrophobic residues often found in incompletely folded proteins. Chaperones, therefore, prevent improper protein interactions, which is especially important under conditions that promote protein unfolding and aggregation, such as environmental stress. Interestingly, this makes them critically important for aging, which has been associated with increased aggregation of disease-related proteins, such as in Huntington’s disease, Alzheimer’s disease, and Parkinson’s disease. The action of chaperones camouflages the adverse effects of polymorphic variants (in the germ line) or accumulated somatic mutations that would normally result in protein-folding defects. Hence, these proteinquality control mechanisms act as effective genetic buffer or capacitor systems against adverse molecular consequences of aging, such as the accumulation of somatic mutations. We have already seen in Chapter 2 that overexpression of heat-shock genes increases the lifespan of nematodes and flies, which illustrates the importance of these systems in aging.
140
GENOME MAINTENANCE
The repertoire of buffering mechanisms is surprisingly diverse and does not only include mutations in the client protein itself, but also in other proteins as well as in generegulatory regions. However, all such proteins become dependent on the chaperone and compromising the function of the capacitor—for example, through a mutation or increased environmental stress—would uncover the hidden mutation load. Although Hsp90 is highly abundant and can be further induced by heat stress, it can be overwhelmed when more and more proteins are destabilized.
4.4 Genome maintenance and aging A key element in current theories of how we age is the accumulation of somatic damage. The decline in the efficacy of natural selection in protecting the soma after the age of first reproduction is the most likely ultimate cause of age-related cellular degeneration and death. Since the first replicators, genome maintenance has been the most important somatic maintenance system, preventing untimely death of its individual carrier and maintaining its genetic integrity, while simultaneously providing the variation on which evolutionary success depends. Together, the various pathways for processing natural damage to DNA and regulating information transfer provide a balance between individual stability and evolutionary diversity. The possible role of genome-maintenance systems in the control of aging and longevity has been recognized almost since the discovery of DNA repair. At that time, the first somatic mutation theory of aging had already been formulated by Leo Szilard392. Alterations in the genome of somatic cells and the possibility that they are a primary causal factor in aging will be discussed extensively in the following chapters. Here I will briefly consider the role of genome maintenance in controlling the rate and extent of such changes and its possible relationship to aging and lifespan.
4.4.1 GENOME MAINTENANCE AS A FUNCTION OF AGE The possible decline of genome maintenance, especially DNA repair, with age and its relation to species-specific lifespan were major topics of aging research, especially during the 1970s and 1980s. Testing these hypotheses was plagued by difficulties encountered in measuring DNA-repair capacity and interpreting the results obtained. From a technical point of view it is clear that there is no easy way to obtain an integral measure for the multitude of DNA-repair activities as they take place in an intact organism. Indeed, even an assessment of one or few DNA-repair activities in a given tissue can be fraught with error. This is simply due to the fact that we know far too little about the mechanisms of repair in intact tissues or cells taken directly from the situation in vivo. The most reliable way of
GENOME MAINTENANCE
141
measuring the capacity of an organism to repair damage in DNA is to analyze the removal of known lesions induced through treatment with relatively low, physiological doses of a genotoxic agent. In my laboratory we have treated rodents with benzo(a)pyrene or N-acetyl-2-aminofluorene and measured the removal of the main lesions from liver and other tissues in young and old animals393,394. Whereas significantly lower levels of removal were observed in the old animals, the difference was only small. Similarly, we have analyzed the capacity of primary rat fibroblasts, derived from young or old animals, to perform DNA-repair synthesis after UV irradiation. Only a very small decline in repair synthesis in cells derived from the old animals was observed395. More recent work has focused on the analysis of BER as a function of age, in view of the importance of this pathway in processing oxidative DNA damage caused by ROS. ROS are now considered as a major contributing factor to the various degenerative aspects of aging. BER activity, measured as the ability of nuclear extracts from various tissues to repair BER substrates, such as uracil, in synthetic duplex DNA, was found by Cabelof et al. to significantly decline with age in various tissues of rodents: brain, liver, spleen, and testes396. This reduction by 50–70% in BER activity correlated with decreased levels and activities of DNA pol , as well as with increased spontaneous mutation frequencies. Similar results were obtained by Intano et al., who specifically looked at germ cells, but in a later study compared brain and liver extracts as well397. Interestingly, these workers found that the age-related BER defect in germ-cell nuclear extracts could be restored by the addition of AP endonuclease, suggesting that this enzyme is rate limiting rather than DNA pol , at least in this tissue398. These investigators have now also analyzed other BER substrates and their results indicate an age-related decline in BER not only using G–U mismatches as substrates, but also for abasic sites. However, they did not observe a decline with 8-oxoG (C. Walter, personal communication). To some extent, these findings are in conflict with results obtained by Imam et al.399, who observed a significant age-related decrease in the in vitro repair of uracil, 8-oxoG and 5-hydroxycytosine in mitochondrial extracts from various regions of the mouse brain. However, in this study nuclear extracts were also tested, without evidence for an age-dependent change in uracil repair399. Using a synthetic DNA duplex with a 1–4 nucleotide gap in one of the strands, evidence has been obtained that neuronal extracts from old rats are deficient in completing gap repair, a measure for BER, possibly due to a deficiency of DNA polymerase and/or DNA ligase400. Interestingly, also using synthetic DNA templates and neuronal extracts, these same workers provided evidence for an age-related decline in NHEJ activity401. Overall, therefore, there is evidence that NER and BER activities and possibly NHEJ activity decline with age, albeit sometimes only marginally and with some discrepancies between the different studies. The possible consequences of the age-related decline in BER activity are still unclear. In all these studies BER activity was measured as glycosylase activity in nuclear or mitochondrial extracts. It should be realized that these assays are difficult to carry out in a reliable, quantitative manner. In addition, glycosylase activities in protein
142
GENOME MAINTENANCE
extracts are not the same as BER activities in cells or in vivo. Even if these results faithfully reflect BER activities in vivo, it is possible that any observed decreases reflect switches in the organism’s utilization of repair pathways or other adaptations to an altered situation driven by the consequences of the aging process. An age-related decrease in cell-proliferative activity comes to mind as a possibility, which may require less BER activity than at a young age. Hanawalt402, Tice and Setlow403, as well as my own group404 reviewed this field in the 1980s and early 1990s, and basically came to the conclusion that there is no evidence for a drastic decline in DNA repair during aging. This is not surprising in view of the critical importance of genome maintenance for the cell. It should be noted that a decline in DNA repair is not necessary to explain an accumulation of alterations in the somatic genome. Since by nature genome-maintenance systems are imperfect, one would expect such alterations to accumulate. However, in the presence of declining repair activities such an accumulation would be expected to be exponential rather than linear. Apart from the absence of a general marker for DNA-repair capacity, it is also very difficult to interpret higher or lower levels of particular repair processes. Genome-maintenance pathways necessarily act in the short-term interest of the cell or the individual. A high activity of a repair pathway could be beneficial for the survival of a cell population affected by genotoxic stress. However, this may be at the cost of reduced genomic integrity, for example, when the repair pathway is error-prone. At old age, therefore, organisms may actually benefit from somewhat reduced repair activities. Along the same lines, an easily triggered DNA-damage response could lead to increased apoptosis rates, which may not be a problem at early age. However, at later age, such increased genome maintenance may cause organ dysfunction due to loss of functional cells. It is because of these considerations that changes in DNA-repair activities are not always easy to interpret. It is possible that DNA-repair activities decline with age as part of a more general decline in enzymatic function. On the other hand, when it is true that aging is associated with increased genotoxic stress one would probably expect an increase in DNA repair in response to the increased damage. Both possibilities may actually be true, as can be illustrated by the apoptosis response, a major component of genome maintenance. In an important series of experiments,Yousin Suh, then in Seoul, South Korea, first observed an association between the resistance of rat liver to tumor formation after treatment with the direct-acting genotoxic agent methyl methanesulfonate and the robust apoptosis response of this tissue405. This was in contrast to the brain, which is highly susceptible to tumor induction by this agent, but displays a lack of apoptotic response. She then tested the apoptotic response in liver to methyl methanesulfonate in rats at old age and found this to be diminished406. Hence, whereas at young age the animals responded to the challenge of this mutagen with a robust apoptosis response, underscoring the importance of this system in getting rid of heavily damaged cells to prevent cancer, at old age this defense system was clearly not functioning well. Without treatment, however, she observed a trend towards an increased level of apoptotic cells in liver of old rodents, which
GENOME MAINTENANCE
143
confirmed earlier observations by others407. Hence, it is possible that a system becomes both less effective at old age in responding to a challenge, but increasingly active during aging due to increased spontaneous genotoxic stress. Subsequent findings by others of a significant reduction in mean apoptotic response to radiation with increasing age in human peripheral blood lymphocytes408 suggests that such a decline may be a general phenomenon associated with the aging process in mammals.
4.4.2 DNA REPAIR IN STEM CELLS Whereas DNA damage in somatic cells can have numerous deleterious consequences, of which malignant transformation and cell death are the most critical, damage accumulation in stem cells would severely constrain the capacity of such cells to replenish organs. Indirect evidence, such as the increased expression levels of DNA-repair genes in stem cells409, suggests that genome-maintenance capacity in stem cells is generally higher than in normal somatic cells. The first direct evidence that stem cells do have increased capabilities to maintain their genome integrity comes from the laboratories of Peter Stambrook (Cincinnati, OH, USA) and Jay Tischfield (Piscataway, NJ, USA). These investigators have been mentioned previously with regard to their discovery that LOH as a consequence of mitotic recombination is a frequent spontaneous event in cells in vivo. Using the Hprt and Aprt selectable marker genes as a measure for mutations (described in more detail in Chapter 6), they compared mouse embryonic stem (ES) cells with mouse embryonic fibroblasts410. The results indicate a dramatically lower spontaneous mutation frequency in the stem cells compared with the embryonic fibroblasts (Fig. 4.8). Mutation frequency 100 000 Aprt
Hprt
Frequency (ⴛ10ⴚ8)
10 000 1000 100 10 1 0.1